Skip to content

RELEASE_NOTES.md

Latest commit

 

History

History
150 lines (125 loc) · 9.01 KB

RELEASE_NOTES.md

File metadata and controls

150 lines (125 loc) · 9.01 KB
Raw

Release Notes

v1.9.15 — Sparkplug B edge node

Event Engine can now act as a Sparkplug B edge node, so an industrial or building-automation host (Ignition, Honeywell EBI, Cirrus Link and other Tahu-compatible systems) can consume device data and send commands back. The usual answer to a Sparkplug requirement is to place an industrial gateway between the Axis device and the Sparkplug network; with this release the device is the edge node itself.

Installation

The signed_*.eap packages are signed with the Axis ACAP signing service and install normally on AXIS OS 12.10 and later.

Upgrading from v1.9.13 or earlier (which were unsigned) can fail with "Couldn't install: app", because the vendor id in the manifest does not match the previously installed version. Back up your rules first (Rules → Export), then uninstall the old app before installing this one. Upgrading from v1.9.14 onwards is a normal in-place install.

Each package is byte-for-byte the built artifact with an armored PGP signature appended, so you can check one against its build:

head -c $(( $(stat -f%z signed_Event_Engine_1_9_15_aarch64.eap) - 849 )) \
    signed_Event_Engine_1_9_15_aarch64.eap | shasum -a 256

What is new in 1.9.15

  • Sparkplug B Edge Node settings section: group / edge node / device / primary host ids, a choice of Sparkplug 3.0 or 2.2, and a central metric list. Metrics can be bound to a device event (published automatically on change) or left unbound and driven from a rule.
  • Sparkplug B Publish action — publish declared metrics as NDATA/DDATA with template-aware values.
  • Sparkplug Command trigger — fires when a host writes a metric via NCMD/DCMD, injecting {{trigger.metric}} and {{trigger.value}}. This is what lets a SCADA system command the device, not just read from it.
  • GET /sparkplug returns the live session state and every metric's current value; the Settings tab renders it as a table.
  • tools/sparkplug_host.py, a dependency-light validation host that decodes the protobuf itself and checks birth-before-data, seq continuity, bdSeq matching and alias resolution.
  • Backup brokers. The MQTT settings take a list of alternate brokers that are tried in order when the primary is unreachable, so a Sparkplug node is not tied to the availability of a single broker.

Protobuf encoding is implemented in-tree, so no new libraries are bundled.

Correctness details

  • bdSeq is persisted across restarts, so a late death certificate from a previous session cannot be mistaken for the current one.
  • The death certificate is registered as the MQTT will and published explicitly on a graceful shutdown, because a clean MQTT disconnect suppresses the will.
  • Sequence numbers are allocated in one place under a lock, so concurrent rules cannot interleave them.
  • Metrics collected while the broker is unreachable are buffered and replayed on reconnect flagged as historical. The same applies while a configured primary host reports itself offline. A node with no primary host, or one whose host has never published STATE, always publishes, so a misconfigured host id cannot silently mute a device.
  • Commands are accepted only on the node's own command topics, and a command delivered twice because a rule subscribes to an overlapping topic filter is acted on once.
  • Both Node Control/Rebirth and Device Control/Rebirth are honoured.

Fixes in 1.9.15

  • MQTT could stop reconnecting permanently after a settings change. Changing the broker host or port closed the worker thread's socket from another thread. select() does not reliably notice that, and once the file descriptor number was reused elsewhere in the process the worker kept operating on an unrelated socket: it never reconnected and could write MQTT bytes into another connection. Reconnects are now requested through a wakeup pipe so the worker always closes its own socket, and a deliberate reconnect no longer waits out the failure backoff. Changing the username, password or client ID now also triggers a reconnect.
  • MQTT: packets larger than the 4 KB receive buffer were parsed past the end of that buffer and left the stream unframed. Oversized packets are now drained safely.
  • MQTT: MQTT_Publish measured payloads with strlen, which truncated any payload containing a NUL byte. Binary publishing is now supported.
  • Turning MQTT off left the broker connection open. The worker only checked the enabled flag before starting a connection, never while one was running, so a disabled client kept pinging and receiving until something else dropped the socket.
  • MQTT: the broker configuration was read by the worker thread while the settings thread overwrote it, which could pair a host from one configuration with the credentials or TLS setting of another. All access now goes through a lock and each connection attempt works from a private snapshot.
  • MQTT: correcting the broker address while the client was backing off could take up to a minute to apply. Retry waits are now interrupted by a settings change.
  • MQTT: a malformed remaining-length field could make the receive loop drain hundreds of megabytes and lose framing. It is now rejected.
  • MQTT: packet assembly ignored allocation failures and could dereference a null pointer under memory pressure.
  • Sparkplug: a failed replay of buffered samples discarded them anyway, losing exactly the data the buffer exists to protect.
  • Deleting a rule left its MQTT topic subscribed. The broker kept delivering on that topic for the life of the app. Subscriptions are now released when a rule is deleted, disabled or edited, and are reference counted so one rule releasing a topic cannot cut off another rule or the Sparkplug command feed.
  • Posting a rule whose body carried an existing id created a second rule with the same id, leaving get, update and delete ambiguous. Importing a backup on top of the rules it came from duplicated every one of them. Such a request now updates the existing rule.
  • Sparkplug: GET /sparkplug reported a metric binding only as the flattened source_path, which the settings endpoint does not accept, so posting that response back silently dropped every binding. The response now also carries the writable source object, and a metric that arrives with source_path but no source is reported by name in the log instead of being quietly ignored.
  • Settings: repeated no-op settings callbacks no longer cycle ACAP event subscriptions.

v1.9.14-Signed - Signed release

  • Packages are now signed with the Axis ACAP signing service and install normally on AXIS OS 12.10 and later.
  • Vendor updated to moshe@mohome.net with the registered vendor ID.
  • Upgrading from an earlier unsigned version can fail with "Couldn't install: app" (device log: "Vendor ID in manifest does not match the vendor ID of the previous version"). Back up your config, uninstall the old version, then install this one.

v1.9.14 — AXIS OS 13 ready

AXIS OS 13 compatibility

  • Rebuilt against the AXIS OS 13 SDK (acap-native-sdk:12.10.0). The manifest was migrated to schema v2 with a compatibleOsVersions declaration of 11.11–13.
  • A single binary still runs on older firmware. Two toolchain symbol-version traps introduced by the newer SDK are pinned so the app loads on older AXIS OS:
    • GLib 2.76's g_string_free_and_steal (capped via the GLib API version in the Makefile).
    • glibc 2.42's cfsetispeed / cfsetospeed (pinned to each architecture's glibc baseline).
    • Verified running on AXIS OS 11.11 and 12.10.
  • SD-card recording was migrated off the removed record.cgi / stop.cgi to the Edge Storage continuous-recording API, with automatic fallback to the legacy CGIs on products that don't support continuous-recording profiles.

New

  • Per-device HTTPS. Actions and conditions that target a remote camera now have a Use HTTPS toggle (remote_https). Required for AXIS OS 13 cameras that are HTTPS-only; the remote device's self-signed certificate is accepted. Leave it off for remotes still served over HTTP, so mixed fleets work either way.
  • Recording stream options. The recording action gains an advanced Stream Options field (options), e.g. resolution=1920x1080&fps=15 or videocodec=h265. Takes precedence over the stream profile. When both are blank it defaults to videocodec=h264 at native resolution (the OS 13 continuous-recording API requires an options value).

Fixes

  • io_state condition now reads the actual io/port.cgi response format (portN=active/portN=inactive); it previously matched active=yes/active=no, which no firmware returns, so the condition always evaluated false.
  • Light Control Flash strobe cache-bust fix (carried from the in-progress 1.9.14).

Notes

  • Production installs on AXIS OS 13 require the .eap to be signed via the Axis ACAP Portal (OS 13 rejects unsigned applications).
  • Both aarch64 and armv7hf builds are symbol-verified; runtime testing was performed on aarch64 (OS 11.11 and 12.10).