edge: support lifecycle event subscriptions and glob device_id filters#27
Merged
Conversation
Lets drivers react to peers coming and going via the existing @on decorator. The registry already publishes device.online / device.offline on a shared per-tenant subject; this patch teaches @on to recognize those event names (plus peer_present / peer_lost aliases) and route them to that subject instead of the per-device .event.<name> pattern. Also adds glob support to device_id=... filters using fnmatch, since NATS/Zenoh wildcards don't span sub-tokens (interlock-* can't be expressed at the broker). Glob patterns subscribe to the broker single-token wildcard and filter post-hoc. @on(device_id="interlock-*", event_name="peer_lost") async def on_any_interlock_lost(self, device_id, event_name, payload): await self.set_power(0.0) Per-device .event.<name> path is unchanged; existing @on subscriptions are byte-identical. Verified end-to-end on both NATS and Zenoh with the new examples/peer-lifecycle/ demo. Closes #26
Collaborator
|
Looks great for registry mode! Quick gap I wanted to flag: the I put up #31 stacked on this branch as a concrete proposal: Happy to fold #31's diff into this PR if you'd rather keep it one merge, or land it separately after this one -- whatever fits your review style. |
kavya-chennoju
pushed a commit
that referenced
this pull request
May 12, 2026
Stacked on #27. Extends peer lifecycle subscriptions to D2D mode. ## What changes #27 routes \`@on(event_name=\"peer_present\"|\"peer_lost\")\` to the registry-published \`device-connect.<tenant>.device.{online,offline}\` subject. D2D mode has no registry publishing those subjects, so D2D drivers using the same decorator see nothing. This PR delivers the same logical events via \`PresenceCollector\` callbacks when a D2D collector is present on the device runtime. The \`@on\` surface, alias names, and \`device_id=\` filter (exact + glob) are unchanged -- mode-transparent behavior. ## How \`discovery.py\`: - \`PresenceCollector\` grows symmetric \`on_peer_removed\` support (fires on \`departing: true\` graceful departure AND on prune-timeout in \`_prune_loop\`). - Multi-listener model: constructor-form \`on_new_peer\` / \`on_peer_removed\` kwargs still work as single-listener seeds; new \`add_on_new_peer\` / \`add_on_peer_removed\` methods append additional listeners. The existing \`collector._on_new_peer = ...\` pattern in \`device.py\` keeps working via a back-compat property. \`drivers/base.py\`: - \`_setup_subscription\` checks \`self._device._d2d_collector\` for lifecycle aliases; if present, the handler is wired via the collector's new listener methods (\`_setup_lifecycle_d2d\`) instead of subscribing to the broker subject. - Per-device events are unaffected; still subscribe through the broker. Glob and exact \`device_id=\` filters apply post-hoc inside the delivery wrapper (same \`_device_id_matches\` helper as #27). ## Tests 9 new tests, 437 total passing (30 from #27 + 7 prior lifecycle + new): \`\`\` TestLifecycleSubscriptionsD2D test_peer_lost_routes_through_collector test_peer_present_routes_through_collector test_d2d_lifecycle_device_id_filter_exact test_d2d_lifecycle_device_id_filter_glob test_d2d_per_device_event_still_uses_broker TestPresenceCollectorRemovedCallback test_graceful_departure_fires_on_peer_removed test_prune_timeout_fires_on_peer_removed test_add_on_peer_removed_supports_multiple_listeners test_constructor_callback_coexists_with_listeners \`\`\` ## Why stacked This PR depends on #27's \`_LIFECYCLE_ALIAS_TO_CANONICAL\` and \`_device_id_matches\` helpers; without #27, the D2D path has no canonical lifecycle name to bind against. Once #27 lands the diff against \`main\` is the 390 LOC here.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #26.
What changes
The registry already publishes
device-connect.<tenant>.device.{online,offline}on a shared per-tenant subject. This PR teaches@onto subscribe to those — drivers can now react to peers appearing and disappearing through the same decorator they already use, without polling or invoke timeouts.Also adds glob support to
device_id=filters: NATS/Zenoh single-token wildcards can't express sub-token patterns likeinterlock-*, so glob patterns subscribe to the broker single-token wildcard and filter post-hoc withfnmatch.Per-device
.event.<name>subscriptions are byte-identical; existing@onusers see no behavior change.Files
device_connect_edge/drivers/base.py— lifecycle subject routing, glob filter,_device_id_matches()helper,_LIFECYCLE_ALIAS_TO_CANONICALmap.tests/test_drivers.py—TestLifecycleSubscriptions: 7 new tests covering subject routing, alias mapping, exact + globdevice_idfilter, per-device regression check, glob-on-per-device-events.examples/peer-lifecycle/device_{a,b}.py— 2-device runnable demo.Tests
End-to-end verification
device_id=filterdevice_id="device-*"(caught 2 peers)NATS run, glob filter:
Known follow-ups (separate PRs)
device_type=filter for lifecycle events — needs ~5 LOC server-side change to includedevice_typein the registry's lifecycle event payload. Skipped here because relying on the D2D peer cache mid-offline is racy.tag=filter — needs a newDeviceIdentity.tagsfield; bigger API change.