# HQbird 2027: What's New

This document lists the changes in each HQbird 2027 release, newest first.
Each release adds a new section at the top; earlier sections stay.

Downloads: <https://ib-aid.com/download/hqbird/v2027/>
(Windows setup `.exe` and Linux `.run` for Firebird 2.5, 3.0, 4.0 and 5.0;
`SHA256SUMS` has the checksums).
Docker images: `ghcr.io/ibsurgeon/hqbird2027-fb25`, `-fb30`, `-fb40`, `-fb50`.

---

## 2027.2.33 (2026-11-16)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 3.4.0 | yes |
| HQbird Monitor | 1.57.0 | yes |
| Config editor (in HQbird Monitor) | 1.61.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.24 | no |
| HQbird RCM | 2027.5.12 | yes |
| HQCluster node | 2027.5.12 | version only, no change in the code |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | no |

### Notifications to Telegram, webhooks and e-mail

- The HQbird agent now sends its own messages about server problems. It
  sends a message when a problem is new, when it gets worse and when it is
  gone. All changes of one status check go in one message.
- The channels are a Telegram bot of your own, a webhook and e-mail (SMTP).
  You can have several channels.
- **HQbird Monitor** has a new page, **Notifications** (the button is on the
  Servers page). On it an administrator can:
  - choose who sends: the agent on this server, the HQCenter, both, or
    nobody (`auto`, `local`, `center`, `both`, `off`);
  - add, change and delete channels;
  - send a test message and find the chat ID of a Telegram bot;
  - see the last 50 deliveries.
- The page never shows a stored token, URL or password. An empty field keeps
  the stored value.
- **Daily summary.** Set `notify.daily_summary` to a time (`"HH:MM"`) and the
  agent sends one message a day with the open problems. It is off by default.
- **Several agents on one machine.** Only one of them sends the problems of
  the machine itself (disk, memory), so a message does not come twice.
- Two new commands of the agent, both without a start of the agent:
  `--notify-history [N]` prints the last deliveries, `--notify-test <channel>`
  sends a test message to one channel.
- The agent does not send while its license does not allow it. It sends one
  last message that says so.
- The config editor (1.61.0) has the section `notify` with the fields
  `notify.mode`, `notify.host_scope`, `notify.classes` and
  `notify.daily_summary`. The channels are not in `agent_config.json`: set
  them on the Notifications page.
- **HQbird RCM** sends its alerts to the same kinds of channels. The channel
  list is the new section `notify.channels` of `rcm.json`. The Alerts page
  has the block **Notification channels** with a **Test** button. RCM sends
  problems of severity `warn` and higher.
- The Notifications page also has a block **HQbird Telegram bot**: alerts
  through a bot of the vendor, with no bot of your own. The vendor's servers
  do not offer this service yet. Until they do, use your own Telegram bot, a
  webhook or e-mail.

Nothing is sent until you add a channel.

### A new license arrives by itself

- An agent that got its license with a trial request keeps a link to the
  vendor's license server. It asks that server once an hour and installs the
  license the server has for it: a longer trial, a paid license made from the
  trial, or the next period of a subscription. Nothing is done on the server.
- A trial request can now be sent after the 30 days of the trial too: the
  e-mail address may belong to a customer with a license.
- The vendor's license servers start to serve such licenses step by step.
  Until then nothing changes for a server.

### Upgrade from HQbird 2024: databases and old logs

When the agent imports the configuration of FBDataGuard (HQbird 2024), it
now does two more things:

- **It registers the databases** of FBDataGuard in the Firebird aliases file
  (`databases.conf`, or `aliases.conf` on Firebird 2.5) as `name = path`
  lines. The old file is copied first. A database is left out when its path
  is already in the file, or when its file does not exist. A file with an
  `include` directive is not changed.
- **It removes the old monitoring logs** of FBDataGuard (the folders
  `mon\outcoming` and `mon\outtrace` of its output folder; the folders
  stay). Nothing is removed while a service of the old HQbird runs. The
  removal cannot be undone.

Both are on by default. To turn them off, set
`hqbird_import.register_databases` or `hqbird_import.clean_mon_logs` to
`false` (the config editor has both fields), or use `--no-register-databases`
and `--no-clean-mon-logs` with `--import-hqbird-config`.

### Docker images

- **A cluster of two containers without HQCenter.** The new setting
  `HQBIRD_CLUSTER_ROLE` of the first start: `master` installs RCM in the
  container, `replica` opens the pairing port. Without the setting a
  container runs no cluster component, as before. See the Docker guide,
  section 7.
- **Fixed: folders open for writing.** In the images up to 2027.2.32 the
  folders `/etc`, `/usr` and `/usr/lib` and some files were writable by every
  account in the container. Use the 2027.2.33 images.
- **Fixed: RCM made in a container did not start in a new container.** The
  container now keeps the service accounts on the `state` volume. A cluster
  made by HQCenter was not affected.

### HQbird RCM 2027.5.12

- Alert notifications (see above).
- Fixed: the page of a forced password change loaded itself again and again.

### Upgrading

- Install or update with the 2027.2.33 Windows setup or Linux `.run`. A
  server with automatic updates takes the agent 3.4.0 and HQbird Monitor
  1.57.0 by itself.
- Notifications need no action: nothing is sent until you add a channel on
  the Notifications page.
- If you upgrade a server from HQbird 2024, read "Upgrade from HQbird 2024"
  above before the import: the import changes the aliases file and removes
  the old monitoring logs unless you turn that off.
- Docker: pull the 2027.2.33 image and start a new container on the same
  volumes.
- A cluster: update HQbird RCM first, then the nodes.


## 2027.2.32 (2026-11-15)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 3.3.1 | yes |
| HQbird Monitor | 1.56.1 | no |
| Config editor (in HQbird Monitor) | 1.60.0 | no |
| fbmcp (MCP server for Firebird) | 2027.2.24 | no |
| HQbird RCM | 2027.5.11 | no |
| HQCluster node | 2027.5.11 | no |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | no |

### Windows setup: Automatic updates page

- The wizard has a new page, **Automatic updates**. It is shown when the
  setup installs a new agent that is attached to no HQCenter. The choices:
  - **agent** (default): the HQbird agent, HQbird Monitor and fbmcp update
    themselves. A new Firebird package is downloaded and waits for the
    administrator.
  - **all**: Firebird is updated too, in its apply window.
  - **off**: the agent makes no update checks.
- For a silent install there is a new switch: `/AUTOUPDATE=all|agent|off`.
  Without the page and without the switch the setup does not change the
  update mode.
- The 2027.2.31 Windows setup did not have this page and this switch. The
  Linux `.run` has had the same choice since 2027.2.31: an item in the text
  interface and the option `--auto-update all|agent|off`.

### Backup pins for the update servers

- The agent holds a pinned key for the update server of each zone. Now it
  also holds a **backup pin** for each zone: the pin of a spare key that is
  kept away from the server. If a zone server must change its key, agents
  with no HQCenter keep their updates and need no reinstall.

### Network: what a server without HQCenter connects to

A server that is attached to no HQCenter takes its updates from a vendor
zone. The agent opens outgoing HTTPS connections to these addresses only:

| Purpose | Address |
|---|---|
| Updates, zone EU | `liceu.ibsurgeon.com:9443` |
| Updates, zone BR | `licbr.ibsurgeon.com:9443` |
| Updates, zone RU | `licru.ibsurgeon.com:20943` |
| Trial license | `trialeu.ibsurgeon.com:9443`, `trialbr.ibsurgeon.com:19443`, `trialru.ibsurgeon.com:23943` |

No incoming port is needed. A server that is attached to an HQCenter does
not use these addresses: it takes its updates from that HQCenter.

To turn the update checks off:

- Windows setup: `/AUTOUPDATE=off`, or **off** on the Automatic updates page.
- Linux `.run`: `--auto-update off`.
- An installed server: the **Updates** page of HQbird Monitor.

### Upgrading

Install or update with the 2027.2.32 Windows setup or Linux `.run`. Use
2027.2.32 and not 2027.2.31 for Windows, because only 2027.2.32 has the
Automatic updates page and the `/AUTOUPDATE=` switch. The agent 3.3.1
replaces 3.3.0 with no change in configuration.



## 2027.2.31 (2026-11-15)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 3.3.0 | yes |
| HQbird Monitor | 1.56.1 | yes |
| Config editor (in HQbird Monitor) | 1.60.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.24 | yes |
| HQbird RCM | 2027.5.11 | yes |
| HQCluster node | 2027.5.11 | yes |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | no |

### Performance report: fuller data, reports on request

- The agent builds richer reports: databases, runs by hour, procedures and
  triggers, who/when/from where, failed runs, and correct attachment counts.
  A scheduled window can wait for late traces and finish as **partial**; a
  hard failure becomes a terminal **error** stub instead of retrying forever.
- **On demand:** any range (with optional database, threshold and process
  filter) from the CLI, Local API, remote command, config editor **Run now**,
  or HQ Monitor **Build report…**. One such report at a time.
- HQ Monitor shows every ranking, compare two windows, 30-day statement
  history, database filter, and the on-demand list. Needs fbagent 3.3.0 and
  config editor 1.60.0 for Build report.

### Updates without HQCenter

- A trial agent (no HQCenter) can take releases from a **vendor zone**
  catalog: agent, Monitor and fbmcp update themselves; Firebird packages
  download and wait for apply. Keys `update.source` / `update.mode`,
  `--update-status`, `--setup-install --auto-update`.

### RCM Services UI (2027.5.11)

- Browser pages for services: master/replicas/hosts/routers, switchover,
  rejoin, resume, cancel, failover, automatic failover rule, Operations,
  Find a database, Settings/Routers, viewer role. Lab-checked. A failed
  service may fail over.

### fbmcp cluster services (2027.2.24)

- Tools that mirror the RCM Services pages (`fb_cluster_service*`,
  switchover/failover/rejoin, policy, routers). Needs RCM 2027.5.11+.

### Upgrading

Install or update with the 2027.2.31 Windows setup or Linux `.run`. Update
node and RCM together to 2027.5.11 before relying on the Services UI or
fbmcp cluster-service tools. On-demand perf reports need agent 3.3.0 and
Monitor 1.56.1 (editor 1.60.0).



## 2027.2.30 (2026-11-14)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 3.2.0 | yes |
| HQbird Monitor | 1.55.0 | yes |
| Config editor (in HQbird Monitor) | 1.60.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.23 | yes |
| HQbird RCM | 2027.5.10 | yes |
| HQCluster node | 2027.5.10 | yes |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | no |

### Server status: clearer findings, full disk

- Finding texts name the disk, folder, database, user, client and statement
  where it helps; hints say what to do; numbers are in readable units.
- **Full disk** (incident 2026-11-13): per-disk used percent with free-space
  floors (20 GiB warning / 5 GiB serious), fast filling (nightly backups are
  not treated as fast filling), `disk_full` from `firebird.log` held for 24 h,
  and one-time log errors shown as events for an hour.
- Windows disks are named by volume **label**, not GUID.

### Firebird port vs Local API / HQ Monitor

- Changing `firebird.port` no longer moves the Local API onto the HQ Monitor
  port. Port pin and conflict checks use the Firebird conf port; conflicts
  are logged and refused.

### Backup compression

- Compressed archive extension works for Zip, gz and **zstd** (not only
  Deflate). The config editor lists zstd first among formats; a new gbak job
  still starts with no compression unless you turn it on.

### HQCluster / RCM 2027.5.10 (switchover)

- Service switchover, fence/quiesce, rejoin with `replica_target`, routers
  (HAProxy route table), automatic failover with witnesses, hung-master
  failover.
- Fence after a crash does **not** wait for the journal (protocol minor 25).
- `applied_sequence` waits until Firebird finishes the current segment.

### HQ Monitor 1.55.0

- File-system findings no longer repeat folders already named in the agent
  text. Tests cover the new disk finding families.

### fbmcp 2027.2.23

- Make RCM through the agent and link per agent row; `fb_cluster_unlink`
  takes a server out through RCM, with clear refusals.

### Upgrading

Install or update with the 2027.2.30 Windows setup or Linux `.run`. Update
node and RCM together to 2027.5.10 before relying on switchover/failover.
Agents pick up the new finding texts and port behaviour on update.


## 2027.2.29 (2026-11-12)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 3.1.0 | yes |
| HQbird Monitor | 1.54.0 | yes |
| Config editor (in HQbird Monitor) | 1.59.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.22 | yes |
| HQbird RCM | 2027.5.9 | yes |
| HQCluster node | 2027.5.9 | yes |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | no |

### Cluster on one host (several agents)

- **Slots.** Each agent on a host gets a slot (0–9) with its own pairing
  port and node ports, so two agents of the same machine can both join a
  cluster without sharing folders or ports.
- **Leave and uninstall.** `fbagent --cluster-leave` (and uninstall) leave
  the cluster cleanly; the RCM owner is protected unless `--force`.
- **Pairing window from HQ Monitor and fbmcp.** Open or close the pairing
  window from the Local API, from **Clusters → Cluster on this server** in
  HQ Monitor, or with fbmcp `fb_agent_cluster_pairing`.
- **Enable RCM on this server** from HQ Monitor (password file path only;
  the password is never shown). Cluster admin through the agent’s
  certificate when this host runs RCM.

### Group backup jobs

- **Continue after an error** is on by default for group gbak, nbackup,
  dump and sweep: one failed database no longer stops the rest.
- Job outcomes list the **failed databases**; the config editor shows a
  summary (succeeded / errors / total time) and a failures table.

### Backup Inventory

- Restore rows carry **DB size** and **Duration** (gbak `-ST TDRW` time).
- The agent records `restored_bytes` and `gbak_seconds` on every gbak
  restore; the config editor shows them under Verified.

### HQCluster node / RCM 2027.5.9

- **Reinit can free space** by removing the old replica copy when asked
  (`remove_old_replica`), with clear job progress while Firebird restarts.
- **Publication check before reinit:** a master that journals no table is
  refused with `publication_off` (and an alert) instead of starting a
  useless nbackup.
- Built with **Go 1.26.11**.

### fbmcp 2027.2.22

- Cluster tools for agent pairing and RCM setup through MCP (no SSH to the
  host for the pairing window).

### Upgrading

Install or update with the 2027.2.29 Windows setup or Linux `.run`. Agents
that already use cluster features pick up slots and the leave/uninstall
behaviour on update; open the pairing window again after the update if you
were mid-join.

## 2027.2.28 (2026-11-12)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 3.0.0 | yes |
| HQbird Monitor | 1.53.1 | yes |
| Config editor (in HQbird Monitor) | 1.56.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.21 | no |
| HQbird RCM | 2027.5.8 | no |
| HQCluster node | 2027.5.8 | no |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | no |

### Agent 3.0.0

- **FBDataGuard import** takes over existing verified and nbackup journals
  into the new jobs, keeps FBDataGuard level names (`naming.noPrefix`), and
  names imported jobs with the database name.
- **Group backups can mirror the source folder tree**
  (`paths.mirrorSourceTree`), including after FBDataGuard import.
- **Create a user in a security database** through the Local API
  (`secdb_user_create`); HQ Monitor’s Security.fdb/Users page uses it.
- Built with **Go 1.26.11** (GO-2026-6859 in `net/http`).
- Video/nbackup calendar fix for the HQbird 2027 explainer; gbak login
  options refused in vbackup jobs with a clear message.

### HQbird Monitor 1.53.1 + config editor 1.56.0

- **Security.fdb/Users** settings page and **Create user** dialog.
- Boxes for “Repeat the source folder tree” and nbackup `naming.noPrefix`.
- Configs page collapses equal consecutive conf copies; Firebird version
  from agent 2.98.0+ DetectVersion records is shown again.

## 2027.2.27 (2026-11-10)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.99.0 | yes |
| HQbird Monitor | 1.53.0 | yes |
| Config editor (in HQbird Monitor) | 1.55.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.21 | no |
| HQbird RCM | 2027.5.8 | no |
| HQCluster node | 2027.5.8 | no |
| Firebird | 2.5.9.27194, 3.0.15.33893, 4.0.8.3332, 5.0.5.1917 | yes (3.0, 4.0, 5.0) |

### Firebird builds

Windows and Linux packages for Firebird **3.0.15.33893**, **4.0.8.3332**
and **5.0.5.1917** replace the previous kit builds. Firebird **2.5.9.27194**
is unchanged.

### Server status on every database

The agent still lists five databases in `databases[]`, but it now judges
**every** database on the host. Databases outside the list appear in one
finding per rule (`…/_others`), with a count and the worst names. Rates
(tx/min, deadlocks, GC) come from the agent's own counters for every
database. Each `active_findings` entry carries `evidence` and `hint`.

### Backup jobs that refuse a setting

For agent **2.99.0** and newer, a new vbackup or nbackup job with a setting
the agent refuses is **saved disabled**, with a clear reason. An enabled
job in the live file is not turned off: Save is blocked. Problems shows
such a job in amber; Run now is off for it. A disabled nbackup job that
conflicts with another job of the same database is a warning, not an
error; a forced run of that disabled job is refused so it cannot break the
enabled chain.

### Resource guard defaults (schema 33)

The default CPU threshold is **30%** (was 20). Persist after confirmed
trips is **3** (was 2). Agent self-health restart confirm cycles is **2**
(was 3). A run the guard cancels is no longer counted as a task failure.

### HQbird Monitor report and servers

The Servers page shows the agent verdict, level and headline in the status
cell. The report header uses the verdict; system panels (Agent, Disks,
Transactions, Replication) show detail tables from the snapshot. Three or
more findings with the same key prefix can merge into one card. API answers
use compact JSON and gzip when large enough.

### Config editor

Run now shows which database will run, with tick boxes for a group job.
Nbackup plan mode uses 24-hour time and shows the agent zone offset. Keep
counts sit beside each level; compression options show only for Deflate.

### Upgrading

Install or update with the 2027.2.27 Windows setup or Linux `.run`, or
take the components from the zone catalog after they are published.
Plan mode and cycle retention still need agent 2.98.0 or newer; saving a
refused backup job disabled needs agent 2.99.0 or newer.

---

## 2027.2.26 (2026-11-10)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.98.0 | yes |
| HQbird Monitor | 1.52.0 | yes |
| Config editor (in HQbird Monitor) | 1.53.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.21 | no |
| HQbird RCM | 2027.5.8 | no |
| HQCluster node | 2027.5.8 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Incremental backups: plan mode

An nbackup job, for one database or for a group, can now run from a
**plan**. Each rule names a level (0 to 4), the days and the time. The
configuration editor shows the rules, a calendar of the coming weeks, the
next runs and a storage forecast. Read-only accounts can open the calendar.

```mermaid
flowchart TD
  rules["Rules: level, days, time"] --> same["Same minute: the lower level runs"]
  same --> keep["Built levels stay built; a retry builds only the missing ones"]
  keep --> cycles["Cycle retention keeps a level 0 with its increments"]
```

- **Same minute, lower level.** When two rules start in the same minute, the
  lower level runs. A plan can also force a level 0 when the last full
  backup is older than a set number of days, and a lower level when an
  increment grows past a share of the database size.
- **A failed higher level does not throw away what was built.** If level 0
  is built and a higher level fails, the built levels count. The retry
  builds only the missing levels. A group retry runs only the failed
  databases.
- **Retention can keep whole cycles.** Policy `cycles` keeps a level 0
  together with its increments, by number of cycles, by days and by a size
  limit. The newest complete cycle is never deleted. Policy `levels` is
  unchanged.
- **Plan mode needs agent 2.98.0 or newer.** The editor refuses to save a
  plan, or cycle retention, for an older agent. An agent older than 2.98.0
  is not offered an update while an enabled job uses plan mode or cycle
  retention. A config push to an agent that is already older, an installer
  downgrade and a restored config are not blocked: an agent before 2.98.0
  makes no backup for a plan job and gives no warning.
- **Simple mode, same minute.** A weekly level 0 that starts in the same
  minute as a daily level 1 now runs. A group job runs a slot once, and
  does not run level 0 again after a level 1 cycle.

### Resource guard

- **The hard memory restart now comes in about 1.5 minutes.** The guard
  looks every 30 seconds, including during its grace and its cooldown, and
  no longer waits for a trip or for the 5-minute server status. This
  restart runs only in `mitigate` mode. In `watch` there is no hard limit.
- **A running backup, restore or SQL script holds that restart** until
  memory reaches 1.5 times the limit. Above that the agent restarts and
  names the jobs in the log. While a restart is blocked, the guard cancels
  a running performance report.
- **A CPU trip lasts until the CPU falls.** Low memory no longer ends it at
  once. A memory trip, in the same way, lasts until the memory falls. Hosts
  with frequent CPU trips will see more paused tasks during a cycle.
- **When one running task holds the memory, the guard cancels that run**
  and returns the memory to the operating system. Other tasks are not
  paused for that. A task the guard switches off stays off until you clear
  it and reload the configuration. While it is off, the agent status shows
  a warning.
- **A memory trip during an open CPU cycle takes over.** Tasks the CPU
  cycle had paused run again, and the memory rules decide.
- **A run cut short by a service stop, an update or a self-restart is no
  longer recorded as a timeout.**
- **`watch` set just after `--bootstrap-config` is kept** at the first
  start. Previously the first start turned it into `mitigate`.

### Performance reports

- **The report timeout now stops the build.** `timeout_seconds` stops at
  the next minute file. A cancelled window writes no report. A window whose
  build is cut short twice (a cancel, the timeout, a crash or a memory
  restart) is closed with a short note, and the next window is built. An
  ordinary stop of the agent is not counted.
- **HQbird Monitor says why there is no report.** Those closed windows are
  listed with a one-line reason. The page opens the latest real report by
  default. A query for such a window says "No Perf report for this time".
- **The upload to HQCenter skips the temporary files** that exist only
  while a report is being built.

### databases.conf and aliases.conf

- **Add database** is a dialog with Alias and Path, in place of two browser
  prompts. A refused entry (duplicate alias, a path already in the file, a
  character the file cannot hold) stays in the dialog with what you typed.
  The new entry is unsaved until Save.
- **Firebird 2.5 opens aliases.conf as text only.** The Form/Text switch is
  hidden, and the Configs page shows aliases.conf snapshots.
- **A missing copy has a clear message.** In remote mode, before the agent
  has sent the file, the page says the copy comes after the agent restarts.
- **Save works on a Linux install when the agent cannot write next to the
  file.** The backup then goes to the agent's `bak/` folder. If the file
  itself cannot be written, the error names the account and the command,
  `sudo chown <account> <path>`. A new Linux setup gives databases.conf
  (aliases.conf on 2.5) to the agent account. An existing install needs the
  `chown` once; see Upgrading.
- **After the agent moves to another HQCenter, or is reinstalled with its
  data folder kept,** the first start sends a fresh copy. The editor is not
  left on "no source" until someone edits the file.
- **If Firebird is not running at start,** the agent retries version
  detection until it succeeds. Start-up does not wait for it.

### Verified backup

- **A group test-restore path without a trailing separator is a folder**
  shared by every database of the group. Two databases no longer race to
  create one file, which made gbak fail with "Access is denied".
- **Two databases with the same file name in different folders do not
  restore into one file.** That test restore is skipped and named; the
  backups are still made.
- **A test restore no longer refuses its own copy** because a client is
  attached to it. The agent drops those attachments and restores again.
  Changing the restore path no longer deletes a file the job did not
  restore.
- **A packed restored database keeps the name you set** inside the archive
  (for example `RESTORE_R.FDB`), not the backup file's stem. The archive
  file name does not change.

### Upgrading

- Run the 2027.2.26 setup over the installed version (Windows `.exe`, or
  `sudo sh hqbird2027-fbNN-linux-2027.2.26.run --yes`). Agents with
  auto-update get 2.98.0 from HQCenter.
- Plan mode and cycle retention need agent 2.98.0. An older agent makes no
  backup for a plan job. Do not point a plan at an agent you still need to
  keep below 2.98.0: while such a job is enabled, that agent is not offered
  the update.
- On an existing Linux install, if Save of databases.conf fails with
  "permission denied", run once
  `sudo chown firebird /opt/firebird/databases.conf`
  (on Firebird 2.5: `sudo chown firebird /opt/firebird/aliases.conf`).
- Firebird, HQbird RCM, HQCluster node and fbmcp are the same builds as in
  2027.2.25.

---

## 2027.2.25 (2026-11-09)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.97.0 | yes |
| HQbird Monitor | 1.51.0 | yes |
| Config editor (in HQbird Monitor) | 1.52.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.21 | no |
| HQbird RCM | 2027.5.8 | no |
| HQCluster node | 2027.5.8 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Agent memory and the resource guard

- **The resource guard acts by default.** Mode is `mitigate` at 4 GiB of
  agent memory (was `watch` at 1 GiB), and the agent restarts itself at
  8 GiB. In `watch` the guard only wrote reports, so a perf-report window
  could hold 18–28 GB with no action. Existing agents move to the new
  defaults once (config schema 31). To keep the old behaviour, set
  `resource_guard.mode` to `watch` after the update.
- **Backups and delivery are never paused or turned off by the guard.**
  The protected list now includes vbackup, nbackup, the packer and
  `send_outcoming`. Before, a memory or CPU trip could stop a backup until
  its job was missed, or stop uploads including the guard's own report.
- **A task the guard turns off shows as Suspended** in the configuration
  editor (`disabled_by: resource_guard`). **Clear & enable** turns it back
  on.

### Perf report memory

- **Statements without a plan are grouped by text with literals replaced.**
  Under `perf_group_by: plan`, an EXECUTE BLOCK sent with new values each
  time was a new entry each time, and `process_perf_report` grew to tens of
  gigabytes on replication hosts. Such statements are now keyed with
  strings and numbers replaced by `?`. Minute files stay readable by older
  and newer agents. HQ Monitor finds the Performance report row through the
  new `group_key`.

### Setup

- **The Server name page can check the server again.** The 2027.2.24
  setup asked the agent for an install plan with `--agent-id`, and the
  agent refused that pair, so the page stopped with "Setup could not check
  this server". 2.97.0 accepts `--agent-id` with `--install-plan`.

### Self-restart limits

- **Daily and cooldown restart limits hold across the restarts they
  limit.** They were counted only in memory, so every new process started
  with a fresh allowance and an agent past the hard memory limit could
  restart every half hour all day. The agent now keeps
  `restart_history.json` across starts.

### Upgrading

- Run the 2027.2.25 setup over the installed version (Windows `.exe`, or
  `sudo sh hqbird2027-fbNN-linux-2027.2.25.run --yes`). Agents with
  auto-update get 2.97.0 from HQCenter.
- After the update, check `resource_guard.mode` if you relied on `watch`
  only. Backups stay protected; other tasks may show Suspended after a
  memory trip — use **Clear & enable** when the host is healthy again.

---


## 2027.2.24 (2026-11-09)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.96.0 | yes |
| HQbird Monitor | 1.50.0 | yes |
| Config editor (in HQbird Monitor) | 1.51.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.21 | yes |
| HQbird RCM | 2027.5.8 | yes |
| HQCluster node | 2027.5.8 | yes |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Cluster: publication before init, space and short names

- **A reinit is refused when the publication is off** (or on with no table
  that has a key). Before, RCM could finish an init and show "Ok" while the
  replica never received a change. The init dialog reads the publication
  state first; the job code is publication_off. Sync the publication, then
  init again.
- **"Remove the old replica copy" frees space for a large reinit.** When
  the replica volume is short, the reinit can remove the previous replica
  database before the copy (both nodes need 2027.5.8). The pair stays
  NEEDS_REINIT until the new image is in place. Without the option, a
  space error names it when it would help.
- **Shorter member names.** RCM names a new member from the agent's
  
ode_name (the server name from the setup) as <name>-master /
  <name>-replica, instead of a long id with the date and -fb50. Link
  hosts shows the Member ID. Members that already exist keep their ids.
- **A table list per database** for the verify and the periodic data check.
  A table that a database does not have is "absent", not a mismatch. The
  data-check settings in RCM apply without a node restart.
- **The first init in the smooth mode works under write load** when the
  archive is still empty (the node waits for writers that began before the
  publication). The false parse_no_observations alert on a new master
  with no published database is gone. Conflicts during the replay after a
  smooth reinit no longer raise 
eplication_conflict_resolved.

### Agent updates and traffic

- **The agent id from the setup is the server name as typed** (no -fb50
  suffix). Folders and Windows services keep their instance names.
- **Less traffic to HQCenter / goafts.** One shared mTLS connection; release
  checks about once an hour while the config poll works; a staged node or
  RCM that is not installed is checked every 6 hours; the version report is
  sent only when it changes. The config editor shows the healthy-poll and
  staged-check intervals.

### fbmcp

- Operator CLI (bmcp ctl) prints one line per command; doctor and
  pprove are clearer. Gated tools check arguments before the gate. The
  tool list hides tools the caller cannot use.

### Upgrading

- Run the 2027.2.24 setup over the installed version (Windows .exe, or
  sudo sh hqbird2027-fbNN-linux-2027.2.24.run --yes). Agents with
  auto-update get 2.96.0 from HQCenter.
- Update the agent, HQbird Monitor, RCM and the nodes together for the
  cluster changes (protocol minor 18). An older node refuses
  
emove_old_replica.
- After the update, sync the publication on every master database before
  the next reinit if it was left off.

---

## 2027.2.23 (2026-11-07)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.95.0 | yes |
| HQbird Monitor | 1.49.0 | yes |
| Config editor (in HQbird Monitor) | 1.50.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.20 | no |
| HQbird RCM | 2027.5.7 | no |
| HQCluster node | 2027.5.7 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Command history of every job

- **The agent keeps the commands it launched.** Every job that runs gbak,
  nbackup, gfix or another tool now has a history: up to 50 launches per job
  and phase. In HQbird Monitor, the config editor's Commands band shows the
  two newest launches under each planned command (time, database, command,
  copy button), "Show more" up to 50, and a "differs from the plan" mark.
- **Test restore band:** the two newest events, each with the backup command
  and the restore command that made and read the file.
- **"Will be created" shows the real test restore path**, also when the
  config leaves the vbackup folders blank.

### vbackup checks

- **gbak `-Y` is refused in every vbackup job.** `-Y` sends gbak's output
  to a file, so the agent's log check failed and the backup was deleted.
  The config editor marks the option, and the agent does not apply a config
  with it. The HQbird 2024 import drops `-Y`.
- **A job that streams to S3 refuses `-SE` in the restore options** too.
- **The test restore replaces its old copy also when clients use it.** The
  agent disconnects the other attachments of the old test database first
  (only in its own test restore or backup folder).
- **New vbackup jobs and imported legacy jobs restore with `-rep`**, so gbak
  overwrites the test database a previous run left.

### Setup

- **Linux: the setup opens its ports in ufw** when firewalld is not active,
  and the uninstall removes those rules.
- **Windows: XNET works on the second Firebird of a host.** A Firebird on a
  port other than 3050 gets `IpcName = FIREBIRD_<port>`.
- Shorter setup output.

### Fixes

- **HQbird Monitor shows the agent version of every Windows server.** On a
  server set up with the default folders (for example
  `C:\HQbirdData\fbagent-fb50`), the agent did not send its status file,
  so HQbird Monitor showed no agent version for it. The agent now sends
  the file from any logs folder.
- databases.conf: `DatabaseAccess` is no longer offered in a per-database
  block (Firebird ignores it there; set it in firebird.conf).
- HQbird Monitor: with "reduced motion" in the system, links inside a page
  jump to their target without smooth scrolling.

### Upgrading

- Run the 2027.2.23 setup over the installed version (Windows `.exe`, or
  `sudo sh hqbird2027-fbNN-linux-2027.2.23.run --yes`). Agents with
  auto-update get 2.95.0 from HQCenter.
- **Check your vbackup jobs for gbak `-Y`:** after the update the agent
  refuses such a job. Remove `-Y` and its file name from the backup and
  restore options.
- A databases.conf block that sets `DatabaseAccess` now fails the check:
  move the setting to firebird.conf.

---

## 2027.2.22 (2026-11-06)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.94.2 | yes |
| HQbird Monitor | 1.48.0 | yes |
| Config editor (in HQbird Monitor) | 1.48.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.20 | no |
| HQbird RCM | 2027.5.7 | no |
| HQCluster node | 2027.5.7 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Linux setup through sudo

- **HQ Monitor now installs when the `.run` setup runs through `sudo`.**
  In 2027.2.21 and before, `sudo sh hqbird2027-fbNN-<ver>.run` installed
  Firebird and the agent, but not HQ Monitor (the setup showed only a
  warning). Now the console runs as the owner of `agent_config.json`
  (`firebird`). When HQ Monitor is not installed, the summary gives no
  console address.
- **On a host installed through sudo with 2027.2.21 or before:** run the
  2027.2.22 `.run` setup again with `sudo sh ... --yes`. It updates the
  installation in place and installs HQ Monitor.
- A failed product helper shows its reason in one line of the setup log.
- **Uninstall:** `--setup-uninstall` also removes the RCM that the setup
  made, and on Windows the uninstall of the last Firebird line removes
  fbmcp.

### Encrypted backup archives

- **A vbackup job can encrypt its zip archive with AES-256** (WinZip AES,
  opens in 7-Zip with the password). Keys are in the new
  `backup_encryption` block of the agent config; in HQ Monitor:
  Settings, "Backup encryption keys", and "Encrypt (AES-256)" in the vbackup
  forms. The agent checks the archive with the key before it deletes the
  `.fbk`. Restore, test restore and Run now read encrypted archives on disk
  and in S3. See `docs/BACKUP_ENCRYPTION.md` of the agent.
- **Escrow:** with `escrow_public_key` set, the config snapshot carries the
  passwords sealed to the owner's key (`--escrow-keygen`, `--escrow-open`).
- **HQbird 2024 import** carries the zip passwords of upload jobs over as
  keys, so the agent restores the archives HQbird 2024 sent.
- **A restore on Linux no longer leaves the unpacked backup readable by
  every user** (folder mode 0750, owner of the target folder).
- An agent older than 2.94.2 writes archives in the clear and cannot read
  an encrypted one: update every agent that restores these backups.

### Backups and sweep

- **vbackup `maxDurationMin` stops only the server gbak of its own job.**
  Before, on Firebird 3 and 4 it also stopped another `gbak -se` of the same
  database; on Firebird 2.5 the cancel failed.
- **Sweep no longer disconnects a running `gbak -se`** on Firebird 2.5, 3
  and 4 (`pre_disconnect` with `keep_process_pattern`), when the client gbak
  runs on the same host as the agent.
- **Backup jobs imported from FBDataGuard get twice the time limits by
  default** (vbackup backup 480 min, restore 960 min; nbackup 48 h), when
  the legacy job file does not set them.

### HQ Monitor

- **Servers page: the Firebird mode from the goafts server list.** A console
  that reads its server list from goafts now shows the Firebird mode; this
  needs goafts with the same fix (it comes with the next site stack).
- **Demo mode keeps the word "Database".** A database file named
  `database.fdb` no longer makes demo mode mask every "Database" label.
- **User Guide for 1.47** (English and pt-BR), with new pictures.

### Security

- The compression libraries of the agent and HQ Monitor are updated
  (brotli 1.2.7, klauspost/compress 1.20.2; GO-2026-6888, GO-2026-6865).

---

## 2027.2.21 (2026-11-05)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.94.1 | yes |
| HQbird Monitor | 1.47.0 | yes |
| Config editor (in HQbird Monitor) | 1.47.0 | yes |
| fbmcp (MCP server for Firebird) | 2027.2.20 | new |
| HQbird RCM | 2027.5.7 | no |
| HQCluster node | 2027.5.7 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

2027.2.20 was not released: its setups had an old agent build.
2027.2.21 has all its changes, with agent 2.94.1.

### fbmcp: one per host

- **The setup now installs fbmcp**, the MCP server for Firebird. There is
  one fbmcp per host; it serves every Firebird line of the host. The setup
  runs `fbmcp ctl hqbird-install`. Options: `--no-fbmcp`,
  `--fbmcp-allow-from`, `--fbmcp-port`, `--fbmcp-dns`, `--fbmcp-ip`.
  A problem in this step is a warning, not a failed install.
- API keys are a separate step: `fbmcp ctl apikey new`.
- **The agent updates fbmcp** like its other products. The agents of one
  host take turns through a host lock.
- On Linux the fbmcp binary belongs to root (`/opt/hqbird/fbmcp`). Before an
  update is put in place, root checks its vendor signature. An unsigned
  build or an older version is refused. On a host installed before, run
  `fbagent --refresh-sudoers` as root once.

### Firebird mode

- **HQbird Monitor shows the mode Firebird runs in** (SuperServer,
  SuperClassic, Classic): in the Servers list after the version, and in the
  report header. When `firebird.conf` sets another mode, a note says that
  it takes effect after a Firebird restart.
- New findings: `engine/server_mode_pending` (info, a `ServerMode` change
  waits for a restart) and `engine/server_mode_conflict` (warning, Classic
  runs with `ServerMode = Super`, so new connections fail).
- Prometheus: `fbagent_firebird_server_mode_info{mode}`.

### databases.conf keeps its comments

- **A push of `databases.conf` from HQbird Monitor no longer removes
  comments**, empty lines and CRLF. The agent ships a byte-exact copy of
  the file, and the push editor opens this copy.
- When the file on the server changed after the editor opened it, the
  agent refuses the push. The page asks you to make the change again on the
  current file.
- With an older agent the editor opens the old snapshot and warns before a
  save that comments will be lost.

### Backups

- **nbackup merges the delta of a crashed run at the next tick.** Before,
  the database stayed in backup lock until the next backup run.
- A streamed backup no longer waits for ever on a dropped network link: an
  S3 call with no progress for 3 minutes fails the upload.
- A multipart upload that stays open after a network break is listed in
  `s3-open-uploads.json`; the next run aborts it.
- vbackup `maxDurationMin` now stops a backup through the Services API too.
- A failed test restore removes its partial database (Windows too).
- A restore from object storage on Firebird 3 uses the Services API: about
  five times faster (1 GB: 22 s against 107 s).
- A config with vbackup errors (a duplicate job id, an enabled job with no
  database) is refused, and the config ack names the error.

### Fixes

- **Metadata tracking reports a change only when the schema changed.**
  Before, every `metadata_tracking` run reported a change, because it
  compared the zip files and not the schema text in them.

- **`--cluster-init` and `--cluster-join` no longer leave the agent
  stopped.** The setup also checks the agent service after `--cluster-init`
  and starts it when it is stopped.
- **Windows setup:** log rotation and the agent log now move with the data
  folder. An agent installed before keeps its config: set
  `log_rotation.logs_path` to `storage.data_path` by hand.
- **HQbird Monitor, Servers page:** right after a new install it waits for
  the first agent data and opens the table by itself.
- **Chart time labels are back** at Last 1h and Last 6h.
- **Config editor:** a file_send target has an "Add field…" list; the
  vbackup group form no longer shows settings that do nothing; the database
  picker shows aliases on Firebird 2.5.
- Report colors for serious, warning and ok findings.
- User Guide for 1.46 (English and pt-BR) with new pictures; many console
  texts translated in all five languages.

### Upgrading

- Update the agent and HQbird Monitor together: the Firebird mode and the
  databases.conf push editor need agent 2.94.1.
- Linux hosts installed before: run `fbagent --refresh-sudoers` as root, so
  the agent can update fbmcp.

---

## 2027.2.19 (2026-11-04)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.93.0 | yes |
| HQbird Monitor | 1.46.0 | yes |
| Config editor (in HQbird Monitor) | 1.46.0 | yes |
| HQbird RCM | 2027.5.7 | yes |
| HQCluster node | 2027.5.7 | yes |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Who uses memory

- **A new "Who uses memory" card in HQbird Monitor.** It is on the
  dashboard under the CPU and disk cards, closed at first. It shows a RAM
  bar (Firebird, the agent, other services, file cache, kernel, free), the
  services (Linux) or programs (Windows) that hold the most memory, and
  memory by kind over the selected time range.
- **Databases in the OS cache** (Linux 6.5 and later): how much of each
  database file the operating system keeps in memory.
- **New finding "Firebird memory pressure"** (Linux): the share of time
  Firebird waited for memory. Warning at 10%, serious at 25%. The hint
  names the service limit, the service that holds the RAM and the
  databases that left the OS cache. The thresholds are in the config
  editor, Server status section.
- **The RAM % on Linux is now total minus available.** Before, memory in
  tmpfs and `/dev/shm` (for example Firebird sort files) was not counted.
  On such hosts the percent is now higher; the thresholds do not change.
  The chart marks the minute of the change with a dashed line.
- Memory by service is on by default. To turn it off, use the config
  editor: System metrics, **Memory by service**.

### Cluster: install without HQCenter

The setup now makes a cluster manager (HQbird RCM) on the host and gives it
a first administrator. Other hosts join it from the RCM console.

```mermaid
flowchart LR
  A["Run setup on host 1<br/>(RCM is created)"] --> B["Open the RCM console<br/>https://host1:7444<br/>log in, set a new password"]
  B --> C["Settings - Link hosts<br/>enter host 2 and host 3"]
  C --> D["Hosts join the cluster<br/>(pairing port 7450)"]
```

- **The setup creates the cluster RCM.** At the end it prints the RCM
  console address, the login `rcmadmin` and the first password. RCM asks
  you to change the password at the first login (new **Password** page).
- **The RCM console uses HTTPS** on port 7444. The login page shows the
  cluster CA hash and lets you download the CA certificate.
- **Link hosts.** In the RCM console, Settings, **Link hosts**: enter the
  hosts where HQbird is installed. RCM invites each agent; an agent accepts
  an invitation for 24 hours after setup (port 7450).
- **Settings, Cluster page** in RCM: the cluster, the agents that joined,
  join tokens and revoke.
- **RCM keeps the cluster programs for the hosts.** The setup copies its
  signed HQCluster node and RCM releases to RCM, and the agents offer theirs.
  HQCluster node and RCM on the hosts update from RCM, not from the
  download server.
- **HQbird Monitor, Clusters page** can work with RCM directly: set
  `rcm_admin_url`, `rcm_admin_user` and `rcm_admin_password_file`.
- The setup opens ports 7443, 7444, 7051 and 7450 in the firewall;
  uninstall closes them. To install without RCM, use `--rcm off`.

### Certificates (gencerts)

- `gencerts` keeps the cluster CA: `-add` signs a certificate for a new
  node, `-renew` extends certificates, `-renew-ca` extends the CA, `-list`
  shows the end dates. Before, a new node or a renewal needed a new CA and a
  restart of every node.
- The CA is now valid for 10 years; node certificates stay at 90 days.

### Linux installer (.run)

- The .run now installs when the distribution package
  `firebirdN-server-core` is present (for example on Ubuntu 26.04 with
  HQbird 2024). This package has only the engine libraries. Before, the .run
  stopped with exit code 7. The package `firebirdN-server` (a distribution
  Firebird server) still stops the install.

### Upgrading

- Update the agent and HQbird Monitor together: the memory card needs agent
  2.93.0.
- An agent that is already in a cluster stays as it is. The new cluster
  install applies to new installations.
- Existing HQCenter clusters keep working as before.

---

## 2027.2.18 (2026-11-04)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.92.0 | yes |
| HQbird Monitor | 1.45.0 | yes |
| Config editor (in HQbird Monitor) | 1.45.0 | yes |
| HQbird RCM | 2027.5.6 | no |
| HQCluster node | 2027.5.6 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Backup-restore cycle and restore

- **gfix goes through the local server.** Every gfix of a restore and of
  the cycle (shutdown, online, header fixes, the shutdown again on
  rollback) opens the database as `localhost/<port>:<path>`. Before, gfix
  opened it through the embedded engine first. While the server held the
  file, this failed at once, and the cycle rolled back without a
  shutdown.
- **Windows: a file being read no longer stops the cycle.** When another
  program (for example a virus scanner) held the cycle's state file, the
  write failed with "Access is denied" and the cycle rolled back a healthy
  database. The agent now waits and tries again.
- **Recovery deletes the `.fbk`.** After the agent was stopped during the
  restore or the swap, the recovery kept the backup file. Now it deletes
  it, as the rollback does.
- **gbak options on one line.** The extra backup and restore options are
  one text field, as on the command line: `-p 16384 -bu 2048`. Before,
  they were a list with one argument per line, and `-P 16384` on one line
  made the agent refuse the cycle. Old configs are converted.
- **Sizes of a small database** show in TB, GB, MB or KB, not
  `0.0 GB (+0 MB)`.

### Jobs

- **A new job does not show a deleted job's history.** A job deleted and
  created again at the same place showed the old job's last command and
  result. Now the history belongs to the job by its id.
- **Config editor: the restore consent box is shown once**, under the
  Restore heading. It is one setting for all restore jobs.

### Upgrading

- Update the agent and HQbird Monitor together. The one-line gbak options
  need agent 2.92.0 (config schema 30); with an older agent the editor
  keeps the list.
- After the update the agent rewrites old option lists into one line. A
  rollback puts back the config saved before the update.

---

## 2027.2.17 (2026-11-03)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.91.0 | yes |
| HQbird Monitor | 1.44.0 | yes |
| Config editor (in HQbird Monitor) | 1.44.0 | yes |
| HQbird RCM | 2027.5.6 | yes |
| HQCluster node | 2027.5.6 | version only (same code as 2027.5.5) |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Database state from the header page

- **The agent reads the state of each database from its header**
  (`gstat -h`): full shutdown, single-user and multi-user maintenance,
  nbackup lock and merge, replica, read only. Every 5 minutes it also
  reads the header of databases that nobody is attached to. A database in
  full shutdown has no attachments, so before it did not show at all.
- **No attach to a database in shutdown.** When the header says shutdown
  or maintenance, the agent does not attach for its metrics. In
  single-user maintenance its attachment could take the one place the DBA
  needs.
- **HQbird Monitor shows state badges** under the database name on the
  dashboard and next to it on the database page: full shutdown (red),
  single-user, multi-user shutdown and nbackup lock (yellow), nbackup
  merge, replica and read only (blue). The tooltip says where the state
  comes from and when. A database in full shutdown is listed again.
- **Sweep interval.** The dashboard list and the database page show the
  automatic sweep interval from the header: "Sweep 20000", or "Sweep off".

### Backup-restore cycle renames the database aside

- **No staging folder and no safe copy any more.** The cycle shuts the
  database down `full`, renames it to `<db>.orig4replace_<date>T<time>`,
  backs it up, restores into `<db>.temp4replace` beside it and renames
  the new file in. Each failure before the swap puts the original back
  online. Free space: two database sizes on one volume (file mode), one in
  pipe mode.
- **A database with a `databases.conf` block** is used through temporary
  aliases with the same block. They are written before the shutdown: when
  the agent cannot write `databases.conf`, the run fails and the database
  stays online.
- **Originals kept: 1 by default** (1 to 10), beside the database.
- **The cycle deletes its `.fbk`** after the swap. A backup whose restore
  failed is kept as `<job>_<date>T<time>.restore-failed.fbk`.
- **Config editor:** the cycle form has the backup folder only; the
  staging, safe copy and swap folders and the shutdown mode are gone.
  The database picker does not offer the cycle's own files.

### Config editor: tooltips and lists

- Every field, section and task has a written description, checked
  against the agent's code, in English, Russian, German, Spanish and
  Portuguese (Brazil). The description comes first, the JSON key last.
- Fields with a fixed set of values are lists: compression and its
  level, `logging.level`, `goafts.mode`, the nbackup and dump group job
  order and checks, and others.
- An imported backup family with all jobs off shows `0 of N on` and
  `all jobs off`, not `gates differ`.

### HQbird Monitor

- **Dashboard layout v2.** "← Servers" is back in the header of every
  server page. "Server status" at the start of the time line opens the
  newest report. Six cards start closed and open with a click; the open
  cards are remembered. From 769 px every chart row has two columns.
  Windows agents have no Load Avg line.
- **Errors page:** **Export CSV** downloads every row with the current
  filters and time range; the date of each error is under its time.

### HQbird RCM: the HQbird Monitor look

- The RCM web UI takes the HQbird Monitor style: header and logo, menu,
  badges, tables, cards, buttons, dialogs and the login page. The fonts
  are embedded: no request leaves the RCM host.
- **Color scheme and density.** Settings → RCM → Appearance: Light, Dark
  or High contrast, and Compact, Normal or Sparse. The choice is shared
  with HQbird Monitor on the same host.

### Fixes

- Disk IO by service on Windows: late ETW events were counted twice
  (up to 28% more bytes than the disk after a burst at the end of a
  minute).

### Upgrading

- Update the agent and HQbird Monitor together. The state badges and the
  sweep interval need agent 2.91.0; older agents show none.
- Backup-restore cycle: `paths.stagingRestoreDir`,
  `paths.safeOriginalDir`, `stagingDir` and `shutdown.mode` are not read
  any more (one warning per job). Check free space beside the database.
  `retention.keepSafeOriginals` unset or 0 now keeps 1 original (was 3).
- Update RCM before the HQCluster nodes.

---

## 2027.2.16 (2026-11-03)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.90.0 | yes |
| HQbird Monitor | 1.43.0 | yes |
| Config editor (in HQbird Monitor) | 1.42.0 | yes |
| HQCluster node, RCM | 2027.5.5 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Disk IO by service: who uses the disk

- **The agent now sees who moves the bytes of each disk.** Each minute it
  splits the reads and writes of each measured disk into Firebird, the
  agent and the busiest other services (up to 8 rows; the rest is
  "other").
  - **Linux:** one row per systemd service (cgroup v2), with up to 3
    program names. A Firebird row is marked "shared" when other programs
    run in the same service.
  - **Windows:** one row per process, from a kernel ETW session of the
    agent.
- **HQbird Monitor, Dashboard: "who uses the disk" table** under the Disk
  IO charts. For the chosen disk it shows, for the 10 minutes before the
  snapshot, each service with its programs, reads and writes per second
  and its share of the disk.
- **Firebird lines of this disk.** When the services are known, the
  Firebird read and write lines on the Disk IO chart show this disk only,
  not all disks. The legend then says "this disk". On Windows only the
  read line: writes through the cache belong to the system.
- **Better disk hints in the server status.** The slow disk and disk stall
  hints name the top disk user, for example "top disk user:
  postgresql@16-main 80% of reads (Firebird 10%)". When slow writes wait
  behind another service's reads, the hint names that reader.
- **Config editor:** Disk IO has the switch `disk_io.by_service` (on by
  default).
- **Linux needs IO accounting.** It is off on a plain Ubuntu 24.04,
  Debian 13, Rocky 9 or Alma 9. Without it the table says why and gives
  the command to turn it on:
  `sudo systemctl set-property <agent unit> IOAccounting=yes`.
  The agent does not turn it on itself.
- **Privacy:** with `disk_io.by_service` and `system_metrics.programs` on,
  the **names** of other services and their programs leave the host. Turn
  `system_metrics.programs` off to keep only the Firebird and agent rows,
  or `disk_io.by_service` off to send no services at all.

```mermaid
flowchart LR
    A["Linux: cgroup v2 io.stat<br/>Windows: kernel ETW"] --> B["diskio line:<br/>io_by_service"]
    B --> C["server_status (10 min):<br/>host.disk_io services"]
    C --> D["HQbird Monitor:<br/>who uses the disk"]
    C --> E["disk latency / stall hints"]
```

### HQbird Monitor: navigation and empty pages

- **Breadcrumb line under the header.** Every page shows its path, for
  example `Servers › SRV › employee › Statistics › Trends`. Each step is a
  link. It replaces the **← Servers** button.
- **Metrics and Clients Versions say why there is no data.** By default
  the agent reads MON$ only for the active databases, up to 10 per server.
  The page now says when the agent does not collect this database, or
  collected it only at other times. A snapshot that the agent could not
  read shows the agent's message and an orange frame instead of a row of
  zeros.
- **Dashboard gstat chips** open the report tab that explains them. The
  chip and the trend tab show the same forecast.
- **Report Backups table** shows the full database path and a new
  **kind** column (`none`, `nbackup_locked`, `stale`, ...).

### Fixes

- **Agent: client versions.** A time-out or a lock error on Firebird 3, 4
  or 5 was written as "Firebird 2.5 has no MON$CLIENT_VERSION" and counted
  as a success. Now it is a warning and the database fails the run.
- **Agent: advanced database stats.** An attach error was counted as a
  success. Now the database fails the run, and HQbird Monitor shows the
  warning on the Metrics page.
- HQbird Monitor: long paths in the report Queries table and on the
  Conflicts page wrap inside their cell.
- HQbird Monitor, demo mode: the folders in a finding key are masked too
  (they often hold the customer's name).
- HQbird Monitor: debug fields with console paths go to admins only.

### Upgrading

- Update the agent and HQbird Monitor together. The "who uses the disk"
  table needs agent 2.90.0.
- On Linux, turn on IO accounting for the agent's unit to see the
  services (command above). Without it the other features work as before.
- `disk_io.by_service` is on when the config does not have it. Nothing
  else to do.

---

## 2027.2.15 (2026-11-02)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.89.0 | yes |
| HQbird Monitor | 1.42.0 | yes |
| Config editor (in HQbird Monitor) | 1.41.0 | yes |
| HQCluster node, RCM | 2027.5.5 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### CPU and memory by program

- **The agent now sees which programs use the host.** Each minute it
  records the CPU and memory of Firebird, of the agent itself and of the
  busiest other programs (top 5 by CPU, top 3 by memory). Processes are
  grouped by program name. Processes that live too short to be named are
  counted as "short-lived processes".
- **HQbird Monitor, Dashboard: new Programs row** under Disk IO. A bar
  shows the host CPU split into Firebird, agent, other programs,
  short-lived processes and idle. A table lists the top programs with
  average and maximum CPU, memory and share of RAM. Click a minute on any
  chart to see the 10 minutes around it. A chart beside the table shows
  the same split over the whole time range.
- **Better hints in the server status.** When Firebird uses less than
  half of the host CPU, the CPU hint names the top program, for example
  "top program: java ~58% of host CPU (Firebird 12%)". The memory hint
  names a program that holds more memory than Firebird.
- **Config editor:** Host CPU/RAM has the switch
  `system_metrics.programs` (on by default).
- **Privacy:** with the switch on, the **names** of other programs leave
  the host. Command lines, paths and users do not. Turn the switch off to
  keep the names on the host.

```mermaid
flowchart LR
    A["Agent: one pass over<br/>the processes per minute"] --> B["cpuram line:<br/>programs block"]
    B --> C["server_status (10 min):<br/>host.programs"]
    C --> D["HQbird Monitor:<br/>Programs row"]
    C --> E["CPU / memory hints"]
```

### Disk IO: Firebird reads and true Firebird writes

- **Firebird reads per second** are now on the Disk IO chart (Linux).
  "Disk reads 100 MB/s, Firebird 29 MB/s" can be read at a glance.
  The server status has the Firebird read share per disk, and a slow-read
  hint says when Firebird reads only a small part of the bytes on a disk.
- **Fixed: Firebird writes above the disk's writes.** Linux counts a page
  as written when it turns dirty, also when its file is deleted before it
  reaches the disk (Firebird sort files and `/tmp/firebird`). HQbird
  Monitor could show "Firebird 21 MB/s" on a disk that wrote 15 MB/s. Now
  the agent subtracts these cancelled writes. HQbird Monitor draws the
  Firebird write line only from the new data; the tooltip shows the part
  dropped before the disk.

### Fixes

- **Disk speed test (perf_test) on a Classic server** without
  `database_dir` failed with "Permission denied": it chose a root-owned
  folder. On Linux it now takes the folder of a database that Firebird
  holds open, and skips folders that the agent or Firebird cannot write
  to. Windows Classic still needs `database_dir`.
- HQbird Monitor: the hint of **Temp files (count)** lists the kinds of
  files behind the count (fb_sort, fb_blob, fb_lock, ...).
- HQbird Monitor: a **report link without a date** opens the day of the
  newest snapshot instead of an error.
- HQbird Monitor **User Guide** (English and pt-BR) covers 1.40 and 1.41
  with new pictures.

### Upgrading

- Update the agent and HQbird Monitor together. The Programs row and the
  Firebird read line need agent 2.89.0.
- Right after the agent update, the Firebird write share is empty for
  10 minutes and the Firebird write rate until the first new minute.
- Existing agent configs get the `system_metrics` section on upgrade.
  Nothing else to do.

---
## 2027.2.14 (2026-11-02)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.88.0 | yes |
| HQbird Monitor | 1.41.0 | yes |
| Config editor (in HQbird Monitor) | 1.40.0 | yes |
| HQCluster node, RCM | 2027.5.5 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Backup jobs keep their service files in the agent's data folder

- **The backup folder now holds only backups.** vbackup and nbackup jobs
  (individual, group, dump) keep the journal, status, reports, run
  records, schedule state and gbak logs in
  `<data_path>/job_state/<config path>/job-<id>/`. Before, these files
  were beside the backups and were named after the database file.
- **Files belong to the job id, not to the database name.** A renamed
  database no longer starts a new journal, so the old backups made under
  the old name are rotated again. An old journal that no job takes over
  is merged into the job's journal when the job alone uses the folder.
- **Reports:** the newest 100 are kept (group jobs: 200).
- **nbackup: one journal per chain**, in the folder of the job that makes
  level 0. An advanced job above level 0 with no level 0 job of the same
  database and folder does not run and reports `W_NB_CHAIN_NO_LEVEL0`.
  Before, it made its own level 0.
- **A job removed from the config:** its state folder is deleted after
  30 days. Its backups stay and show as unattributed.
- **Test restore file** is named by the job id:
  `restore.<job id>.fdb` (group: `restore.<job id>.<db key>.fdb`).
- **Config editor:** the job id row shows the folder of the service
  files. A saved job's id is read-only. The journal path field is
  removed; `paths.journalPath` is deprecated (`W_BACKUP_JOURNALPATH_IGNORED`).

```mermaid
flowchart LR
    subgraph Before
      B1[Backup folder] --- B2[backups]
      B1 --- B3["journal, status, reports,<br/>gbak logs (by database name)"]
    end
    subgraph Now
      N1[Backup folder] --- N2[backups only]
      N3["data_path/job_state/.../job-ID"] --- N4["journal, status, reports,<br/>gbak logs (by job id)"]
    end
    Before -->|first run after the update| Now
```

### Disk speed test (perf_test) runs only at night

- **Execution window**, default Saturday 22:00 to Sunday 06:00 in the
  agent's time zone. Outside the window a run is skipped. Inside it the
  test stops before the window ends.
- The default schedule is Sunday 01:00 (was 05:30) with a 900 s limit
  (was no limit). Your own schedule or limit is kept.
- Only one test runs at a time. After a timeout or cancel, the test
  database is dropped (the agent retries for up to 100 s).
- The test no longer writes an error into the Firebird trace.
- The config editor shows the window on the Jobs row; the timeout column
  reads "Timeout, s".

### Fixes

- **Log rotation no longer deletes the agent's state and backups.** With
  the default settings, rotation walked the agent's data folder and
  deleted files older than 10 days, including backups in the default
  backup folder, journals and the license markers.
- **Backups and databases are no longer sent to the console** with the
  logs (fblogspacker 1.5.8). Job reports and gbak logs still travel.
- **Server status counts group backups per database.** A database backed
  up only by a group job had no last backup and a false warning
  `backups/none`.
- HQbird Monitor: **Jobs** and **Backups** are marked in the top menu on
  their own pages. Info messages stay 5 s.

### Upgrading

- Update the agent and HQbird Monitor together.
- The agent moves the service files of each backup job at its first run
  after the update. Until that run, the files are copied, so a rollback
  to 2.87.0 still finds them.
- Remove `paths.journalPath` from backup jobs; the config editor does it
  on Save after the move.
- Check advanced nbackup jobs above level 0: each needs a level 0 (or
  simple) job of the same database and folder.

---

## 2027.2.13 (2026-11-02)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.87.0 | yes |
| HQbird Monitor | 1.40.0 | yes |
| Config editor (in HQbird Monitor) | 1.39.0 | yes |
| HQCluster node, RCM | 2027.5.5 | yes (node fixes; RCM has the same code as 2027.5.4) |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Disk speed of the database and temp disks

- **New job `collect_disk_io`** (Windows and Linux, every minute, on by
  default). It reads the disk counters of the operating system for the
  disks with databases, the Firebird temp folders and
  `disk_io.extra_paths`: operations, bytes, busy time, read and write
  latency. It puts no load on the disk. `disk_io.enabled` turns it off.
- **Server status (schema 22), area io:**
  - `io/disk_latency/<disk>`: reads against 20 / 50 ms, writes against
    10 / 30 ms (with ForcedWrites each commit waits for a write);
  - `io/disk_stall/<disk>`: the slowest minute of the last hour against
    200 / 1000 ms. Two slow writes in one minute are now seen; the
    10-minute average hid them;
  - a disk without databases (temp only) is a warning at most;
  - when the agent's own backup or sweep makes the disk slow, the
    finding is info and names the task.
- **Baselines:** the read and write latency of each database disk joins
  the hour-of-day baselines. 8 ms on a disk that is usually at 0.5 ms is
  reported.
- **Prometheus:** `fbagent_host_disk_reads_total`, `_writes_total`,
  `_read_bytes_total`, `_written_bytes_total`, `_read_time_seconds_total`,
  `_write_time_seconds_total`, `_busy_time_seconds_total` and
  `fbagent_host_disk_info`.
- **HQbird Monitor dashboard:** a new row of Disk IO charts per disk:
  read and write latency with the IOPS, and MB/s read and written. A
  yellow band marks the minutes of the agent's own heavy tasks (backup,
  restore, sweep, disk speed test). On Linux a dashed line shows what
  Firebird wrote: much less than the disk wrote means another program
  loads the disk. The temp files chart has a third line: the speed of
  the temp disks.
- **Config editor:** a **Disk IO** row in the Health check group and the
  new thresholds `disk_io_write_latency_ms` and `disk_io_stall_ms`.

```mermaid
flowchart LR
    K["Disk counters of the OS<br/>(each minute, no extra load)"] --> J[collect_disk_io<br/>database and temp disks]
    J --> S["Server status<br/>disk latency, disk stall"]
    J --> P[Prometheus counters]
    J --> D[HQbird Monitor<br/>Disk IO charts]
    S --> H[Health report<br/>and server strip]
```

### Server state over a period (HQbird Monitor)

- **Health report matrix.** The top of the report shows one row for the
  server and one per system (18 systems). Each cell is 1 h, 4 h or 6 h of
  24 hours, 7 days or 14 days, colored by the worst level in it. A click
  on a cell opens the report at that time. Option "only systems with
  problems". Below: totals, a chronology of the problems, and **System
  details** with charts (CPU, memory, disk write latency, attachments)
  and dashed threshold lines.
- **Server strip** on the Servers page and the dashboard: one cell per
  status snapshot, green, orange or red; hatched grey for a gap. The tip
  shows the time, the level, the headline and the problem areas. A click
  opens the report at that snapshot. On the dashboard it replaces the AI
  report widget.
- **Previous and next snapshot** buttons in the report, also across
  midnight. Back and Forward of the browser now work in the report.

### Backup status

- **Another server's files in a shared backup folder no longer count as
  this server's.** They are shown as *unattributed* files and folders,
  with the info finding `backups/foreign_files`. Its action in the config
  editor opens the backup scan paths.
- **The backup scan has a time limit:** `backup_status.scan_max_seconds`
  (default 120 s). The status names each folder that was cut short.

### Security (Linux installation)

- Root no longer follows symbolic links in the agent and HQbird folders,
  during `--install` and in the installer scripts.
- Root runs only the checked copy of the HQbird agent, from its own
  temp folder.
- Archive extraction (Firebird update, setup) cannot write outside the
  target folder.
- Downloads use https only, redirects included. A host whose curl
  cannot check the certificate pin (CentOS 7) stops with a message: pass
  `--bootstrap-ca FILE`.
- The enrollment token and the SYSDBA password are no longer visible in
  the process list. New options `--token-file FILE` and
  `--sysdba-pass-file FILE`; the environment variables
  `FBAGENT_ENROLL_TOKEN` and `SYSDBA_PASS` also work.
- HQCluster node: no setting can add a line to `replication.conf`
  through a line break, and a database path with a control character is
  skipped.

### Fixes

- `TempDirectories` in firebird.conf is split on `;`, as Firebird does.
- The agent's `bak/` folder keeps the 3 newest copies of each file after
  a self-update. One host had 27 old binaries (1.2 GB) there.
- Linux `--install --service-name X` no longer replaces the default
  agent service.
- HQCluster node: the enroll reply could read settings that another
  request was changing at the same moment.
- HQbird Monitor local install scripts keep the 3 newest binary copies.

### Upgrading

- Update the agent and HQbird Monitor together. The Disk IO charts, the
  disk thresholds and the disk findings need agent 2.87.0; with an older
  agent HQbird Monitor shows no Disk IO row.
- The disk IO job is on after the update. Set `disk_io.enabled` to
  false to turn it off.
- Where a shared backup folder holds other servers' backups, the number
  of databases in the backup status can drop after the update. This is
  expected.
- HQCluster: update RCM first, then the nodes, to the same version
  2027.5.5.

---

## 2027.2.12 (2026-10-31)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.86.0 | yes |
| HQbird Monitor | 1.39.0 | yes |
| Config editor (in HQbird Monitor) | 1.37.0 | yes |
| HQCluster node, RCM | 2027.5.4 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Linux limits of Firebird

On Linux the HQbird agent now watches the operating system limits that
stop Firebird from starting new processes, threads or files. This matters
most for Classic Server, where each connection is a process.

- **What is measured, each minute:** the systemd task limit of the
  Firebird service (TasksMax) and the refused forks, the process limit of
  the Firebird user (LimitNPROC), open files (LimitNOFILE), the host's
  `kernel.threads-max`, `pid_max`, `fs.file-nr` and OOM kills. cgroup v1,
  v2 and hybrid are supported.
- **Server status (schema 19)** has a new area **limits** with the peak of
  the last 10 minutes. A warning comes at 80% of a limit and a serious
  finding at 95% (setting `server_status.thresholds.limits_usage_pct`).
  Refused forks, thread-creation errors in firebird.log and OOM kills are
  serious at once. For Classic the status says how many more connections
  the closest limit allows. The hint gives the command to raise the limit;
  the agent changes nothing itself.
- **Also new in server status:** swap traffic, tasks blocked on I/O and
  CPU steal (finding `memory/swap_activity`).
- **Prometheus:** `fbagent_host_limit_used`, `fbagent_host_limit_max` and
  `fbagent_host_firebird_connections`.
- **Less load:** one pass over `/proc` replaces three. On a host with
  3,500 processes it takes about 17 ms instead of about 100 ms.
- **firebird.log:** thread-limit errors (`pthread_create failed` and
  similar) count as resource errors, not as authentication failures.
  Deleted lock files ("TouchFile failed ... No such file or directory")
  are a new serious class, `lock_files`.
- HQbird Monitor shows the **limits** area in the health report after
  **memory**.

```mermaid
flowchart LR
    P["/proc and /sys<br/>(one pass a minute)"] --> L[Limits per Firebird server<br/>and for the host]
    L --> S["Server status: area limits<br/>80% warning, 95% serious"]
    L --> M[Prometheus gauges]
    F[firebird.log<br/>thread-creation errors] --> S
    S --> H[HQbird Monitor health report]
```

### Backup-restore cycle

- **Saturday 02:00 by default.** A backup-restore cycle job with an empty
  schedule ran every 5 minutes and replaced the live database each time.
  It now runs on Saturday at 02:00 in the agent's time zone
  (`0 0 2 * * 6`). A new job gets this schedule and starts disabled. The
  job form has its own **Schedule** field and shows this default.
- **An interrupted cycle is undone or finished.** The agent keeps a file
  that names the step in progress. After a stop or a crash it reads the
  file at start: a cycle stopped before the swap is rolled back, one
  stopped in the middle of the swap gets the original database back, and
  one stopped after the swap is finished. The job result is a warning with
  the reason `interrupted`. A cycle that fails while the agent runs rolls
  back the same way. Guide: `docs/BACKUP_RESTORE_CYCLE.md` in the agent.

```mermaid
flowchart TD
    A[Agent starts or minute check] --> B{Intent file<br/>for the job?}
    B -- no --> Z[Nothing to do]
    B -- yes --> C{Step in progress}
    C -- before the swap --> R[Roll back: remove staging,<br/>bring the original online]
    C -- between the two renames --> O[Put the original back]
    C -- after the swap --> F[Finish the cycle]
    C -- unknown --> N[Change nothing:<br/>operator needed]
```

### HQbird Monitor and Config editor

- **"Configuration" is now "Jobs"** in the server menu, the server list
  and the editor header. The address `/agent-config/` does not change.
- **Job forms show their fields in titled groups**, in two columns. Rare
  groups are folded under **More settings**; **Expand all** and
  **Collapse all** do every group. A group with an empty required field
  or a warning opens by itself. Groups about one feature (S3 streaming,
  file transfer, test restore, shutdown before the swap, disconnect before
  sweep) have its switch in the title. No field was removed.
- **Sizes next to percents.** The CPU and RAM chart (dashboard and
  Servers page) shows used and total RAM in the RAM % tooltip, for example
  `RAM %: 63 (10.1 GB / 16.0 GB)`. The disk bars show used and free sizes
  under the percents; a tooltip on a bar gives the mount, used, free and
  disk size. The health report CPU card adds the RAM sizes. An older agent
  without sizes still shows only percents.
- On the Servers page the charts of an opened server list all lines of
  the hovered minute (CPU, Load and RAM), as on the dashboard.

### Fixes

- Backup-restore cycle: the free space check adds up the backup file, the
  staging database and the safe copy on each disk they share.
- Backup-restore cycle: the safe-copy rotation of one job no longer counts
  (and deletes) the copies of another job in the same folder.
- Backup-restore cycle: a reconnecting client no longer fails the cycle.
  When Firebird refuses `gbak -b` or `gfix -online` because the database is
  shut down, the step runs again (up to 10 times). Before, a refused
  `gfix -online` could leave the database offline.

### Upgrading

- Update the agent and HQbird Monitor together. An older agent runs a
  backup-restore cycle with an empty schedule on every tick, whatever the
  editor shows. Set an explicit schedule on such jobs if you keep an older
  agent.
- The Linux limits need no setup. Linux only; Windows hosts show no
  **limits** area.

---

## 2027.2.10 (2026-10-31)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.85.0 | yes |
| HQbird Monitor | 1.36.0 | yes |
| Config editor (in HQbird Monitor) | 1.35.0 | yes |
| HQCluster node, RCM | 2027.5.4 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Backups

- **Group backups: "Auto-exclude individual backups".** The vbackup,
  nbackup and nbackup dump group jobs have one new setting. On (the
  default, also for existing configs): a database that has its own
  individual backup job is left to that job. Off: the group job backs it up
  too. For nbackup, "off" gives a warning, because two level chains of one
  database break each other. The old "Autoexclude VIP databases" setting is
  gone from the group forms.
- **Group scans no longer skip databases in use.** A backup does not depend
  on whether a database has connections. A test restore still never writes
  over a database in use.
- **The job result shows the test restore.** For a vbackup job with the test
  restore on, the job result has the restore result (ok, failed, skipped,
  not run), the gbak restore command, the exit code, the duration, and the
  source and target files. HQbird Monitor shows this as a **Test restore**
  row in **Last outcome**.

Which job backs up a database:

```mermaid
flowchart TD
    A[Database found by the group scan] --> B{Has its own<br/>individual job?}
    B -- no --> G[Group job backs it up]
    B -- yes --> C{Auto-exclude individual<br/>backups is on?}
    C -- yes --> I[Individual job backs it up<br/>group job skips it]
    C -- no --> D[Both jobs back it up]
    D --> W{nbackup?}
    W -- yes --> X[Warning W_NB_INDIVIDUAL_INCLUDED:<br/>two level chains of one database]
    W -- no --> G2[OK]
```

### HQbird Monitor and Config editor

- **Top bar buttons say what they do:** **Discard changes**, **Save** and
  **Save and apply now**. When the agent offers "Save and apply now", it is
  the main button.
- **Mask hints.** Each group form says what the include and exclude masks
  are matched against, and what an empty include list means.
- **"+ Add job" scrolls to the new job's form.**

### Fixes

- A database with its own nbackup dump job was also taken by the nbackup
  group scan, so two jobs wrote levels into one backup history.
- A test restore into a folder (for example `C:\TEMP\GROUP RESTORE\`)
  failed. A path that ends with `\` or `/`, or names a folder, now gets a
  default file name: `restore.<db.key>.fdb` (group) or `restore.<db.id>.fdb`
  (individual).
- The backup journal held `null` instead of an empty list after the first
  backup of a database failed.

### Upgrading

- FBDataGuard import: `autoExcludeVIPDB` maps to the new group setting.
- Update the agent and HQbird Monitor together. With an older agent the new
  "Auto-exclude individual backups" box has no effect, and the Test restore
  row is not shown.

---

## 2027.2.9 (2026-10-30)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.84.0 | yes |
| HQbird Monitor | 1.35.1 | yes |
| HQCluster node, RCM | 2027.5.4 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Setup

- **New names by Firebird line and instance number.** A new agent is named
  after the Firebird line, not the port: `fbagent-fb50`, then
  `fbagent-fb50-2`, `fbagent-fb50-3`. The same suffix is on the agent
  service, the HQbird Monitor service, the agent id and the data folder.
  A second Windows Firebird of the line gets `Firebird50-2` and service
  `FirebirdServerHQBirdInstanceFB5-2`. Installed agents keep their names
  on update and repair.
- **Windows setup opens the HQbird Monitor port in Windows Firewall**
  (rule `HQbird Monitor (TCP <port>)`). Uninstall removes the rule.
- **New page order and plain texts** in the Windows setup: Server check
  results, Installation type, Older HQbird services, Installation folders,
  Firebird server, SYSDBA, HQbird Monitor, Trial period, License file,
  Connect to HQCenter, Ready to Install.
- A refused install plan stops **Next** only on the page with the setting
  to change, and says which setting (Firebird port, Firebird folder or
  service name, HQbird Monitor port).
- **Docker:** the container keeps the agent's folder in `setup-state.env`,
  so a changed `HQBIRD_FB_PORT` no longer loses the agent.

```mermaid
flowchart LR
    S1[First fb50 install] --> N1[fbagent-fb50<br/>Firebird50]
    S2[Install another instance] --> N2[fbagent-fb50-2<br/>Firebird50-2]
    S3[One more] --> N3[fbagent-fb50-3<br/>Firebird50-3]
```

### HQbird agent and HQbird Monitor

- The server status gives the time of the daily filesystem check.
- HQbird Monitor server report: the filesystem cards list the paths found
  (up to 5, with the reason) and the time of the check.
- HQbird Monitor server report: larger text.

---

## 2027.2.8 (2026-10-30)

### Versions

| Component | Version | Changed |
|---|---|---|
| HQbird agent (fbagent, enroll) | 2.83.0 | yes |
| HQbird Monitor | 1.35.0 | yes |
| Config editor (in HQbird Monitor) | 1.34.0 | yes |
| HQCluster node, RCM | 2027.5.4 | no |
| Firebird | 2.5.9.27194, 3.0.15.33892, 4.0.8.3331, 5.0.5.1916 | no |

### Setup

- The uninstaller and a copy of the setup go into
  `<HQbird folder>\setup\fbNN`, not into Program Files.
- A silent new install without `/HQMPASS` gets HQbird Monitor user `admin`
  with a generated password, as documented. Before, it ended with exit 1.
- Windows: a new HQCluster node or RCM goes into the HQbird folder
  (`<HQbird folder>\hqclusternode`, `<HQbird folder>\hqbirdrcm`).
  Installed nodes do not move.
- Windows installers work in PowerShell 7.

### HQbird Monitor

- **Server report:** one Errors card and an Errors section with the three
  most frequent kinds of trace error. KPI cards link to their section.
  Top queries show the execution plan and a link to the Performance report.
- **Errors page** shows the application, client address and `ERROR AT`
  function of each trace error. A click puts the value into the filter.
- **databases.conf editor:** one backup copy per edit session instead of
  one per Save, a Reload button, and the editor fills the window. The
  database picker lists the databases and aliases that databases.conf names.
- **Logs:** a server without firebird.log or replication.log shows an empty
  log instead of an error.
- **Demo mode** no longer freezes large pages.
- **Settings** keeps the server navigation.

---

## Earlier releases

See the product changelogs: fbagent `CHANGELOG.md`, HQbird Monitor
`CHANGELOG.md`, Config editor `configeditor/CHANGELOG.md`.
