Repository navigation
Conversation
The plugin targets Toolbox API 1.13.87111 instead of the required 1.15.88839 release. This commit updates the shared API version so the plugin dependencies and packaged manifest consistently target 1.15.88839.
The plugin has no setting for a connection log buffer limit, so users cannot configure how many diagnostic entries a buffer should retain. This commit adds a General settings field backed by the connectionLogBufferSize key. The settings store defaults to 1000 entries and accepts whole numbers from 0 through 10000, with 0 representing disabled buffering. Invalid stored values fall back to the default, while the settings page rejects invalid input before saving. The README describes the setting, and tests cover persistence, range validation, default handling, and reloading the field.
The plugin settings store exposes current values without a way for other components to observe changes. This commit adds typed observation that emits the current effective value and subsequent distinct values. Subscribers can cancel collection independently, and slow collectors receive the latest state. The settings store serializes writes and observer reads, publishing changes after successful storage updates. TLS settings use value equality to suppress duplicate notifications. Existing consumers retain their current behavior. Tests cover initial values, duplicate suppression, failed writes, cancellation, multiple subscribers, and concurrent updates.
Currently, Toolbox discards plugin logs below the active log level, which removes detail needed to diagnose connection failures. This commit adds a private flight recorder to CoderLogger for messages supplied as strings. The recorder uses Toolbox's filtering decision to retain each message with its original level, timestamp, and exception details in a bounded ring buffer. It observes the plugin's buffer capacity setting, preserves the newest records when shrinking, and clears the buffer when recording is disabled.
Currently, CoderLogger delegates lambda logging overloads directly to Toolbox, bypassing the flight recorder. The plugin does not use these overloads yet, but introducing a call later can silently lose filtered messages. This commit overrides every lambda logging overload in CoderLogger, including exception variants, so future calls participate in recording. FlightRecorder evaluates captured messages once and reuses their text, while leaving the original suppliers unevaluated when capture is unnecessary.
Currently, Toolbox disables every log level when logging is off, so the flight recorder treats those messages as filtered and retains them. This commit makes FlightRecorder check Toolbox's effective log level before capturing either string or lambda messages. When the level is off, the recorder leaves buffered entries untouched and does not evaluate message suppliers.
The flight recorder retains suppressed log entries but has no way to replay them into the Toolbox log, so the captured context cannot be written on demand. CoderLogger now exposes a flush operation, and FlightRecorder replays entries from oldest to newest in batches of up to 100. The replay preserves original UTC timestamps and severity in each message, uses a level the Toolbox logger currently persists, and removes entries after their batch is logged. Logging at OFF leaves buffered entries untouched and produces no replay output. The CoderLogger test suite exercises replay formatting, batching, log levels, buffer resizing, and lazy messages through controllable mocked dependencies.
Connection lifecycle messages do not replay the flight recorder, so the Toolbox log can omit the suppressed context leading up to a disconnect or connection-affecting failure. The remote environment and provider now flush after the selected lifecycle messages. The REST client, workspace filter, and action failure handler also flush at their corresponding triggers. Each flush follows its log message so filtered trigger messages appear in the replay, and environment-specific flushes retain the available session ID. Replay tests cover action failure context and filtered status updates through CoderLogger. The REST tests configure the Toolbox logger's effective level and filter explicitly so the new flush calls use realistic logging behavior.
Support bundle export leaves suppressed log entries in memory, so exported Toolbox logs can miss the context recorded before collection. CoderSupportBundleCollector now flushes the flight recorder through CoderLogger before collecting diagnostics. The replay uses the support bundle reason and current session ID to preserve the connection context in the log. The regression test exercises a real CoderLogger and verifies that buffered entries are replayed before bundle collection and removed afterwards.
The README describes the buffer limit but does not explain when recorded logs are replayed or where to find them, leaving connection diagnostics difficult to collect and interpret. The flight recorder documentation now covers configurable capacity, live resizing, log levels, replay triggers and formatting, and Toolbox log locations. The settings instructions identify the action menu in the Coder page header. The collection guidance also explains that per-environment log collection requests a recorder flush and Coder support bundle, while general Toolbox diagnostics collection does not explicitly request either.
The release notes omit the flight recorder and list an outdated Toolbox minimum, while the support bundle requirements do not identify both Coder components. The README and changelog now identify Toolbox 3.9 as the minimum supported version and specify Coder server and CLI 2.10.0 or newer for support bundle collection. The release notes also summarize the flight recorder and explain that Toolbox logs and eligible local SSH logs remain available when a Coder support bundle cannot be generated.
The connection log buffer description wraps across several lines in Toolbox's narrow settings window. The settings page now uses a 46-character description that names the flight recorder and retains the guidance to set its capacity to zero to disable it. The localization catalog uses the same text.
fioan89
marked this pull request as ready for review
October 22, 2026 19:52
fioan89
requested review from
EhabY,
aqandrew,
code-asher,
jeremyruppel and
matifali
October 22, 2026 19:52
CoderRemoteEnvironment logs each action-menu refresh with an unreadable status object, which adds noise to the flight recorder. This commit removes that debug message because the existing status-transition logs already provide readable workspace and agent details.
CoderRemoteEnvironment flushes the flight recorder after every status update, so routine updates drain diagnostic context even when reachability does not change. This commit limits the status-update flush to transitions from reachable to unreachable using the state published to Toolbox. Status messages continue to be logged for every update, while buffered context remains available through other transitions. Regression tests cover all four reachability transition combinations, and the README describes the narrower trigger.
CoderRemoteProvider flushes the flight recorder whenever the environment list changes, so initial discovery and routine additions drain diagnostic context. This commit removes that flush while preserving SSH reconfiguration. Removed environments with an SSH session continue to flush during disposal, and the README now reflects the remaining replay triggers.
There was a problem hiding this comment.
🟡 Changes recommended
Automatic replay can persist sensitive process arguments, while concurrent capture ordering and initial setup-failure replay remain unresolved.
3 open issues
What changed in this PR
Adds a configurable flight recorder to preserve filtered plugin logs and replay diagnostic context during connection lifecycle events, failures, and support-bundle collection.
Changes:
- Adds bounded log buffering, batched replay, and live capacity updates through observable settings.
- Integrates replay triggers and adds recorder, settings, and export tests.
- Documents diagnostic behavior and raises the minimum supported Toolbox version to 3.9.
| File | Description |
|---|---|
| src/test/kotlin/com/coder/toolbox/views/CoderSettingsPageTest.kt | Tests buffer-size validation, saving, and reloading. |
| src/test/kotlin/com/coder/toolbox/views/ActionTest.kt | Updates logger construction. |
| src/test/kotlin/com/coder/toolbox/util/ConnectionMonitoringServiceTest.kt | Supplies logger settings. |
| src/test/kotlin/com/coder/toolbox/util/CoderProtocolHandlerTest.kt | Supplies logger settings. |
| src/test/kotlin/com/coder/toolbox/store/CoderSettingsStoreTest.kt | Tests buffer-size defaults and persistence. |
| src/test/kotlin/com/coder/toolbox/store/CoderSettingsObservationTest.kt | Tests settings observation and concurrent writes. |
| src/test/kotlin/com/coder/toolbox/session/SessionIdRegistryTest.kt | Updates logger settings and coroutine scope. |
| src/test/kotlin/com/coder/toolbox/sdk/CoderRestClientTest.kt | Models effective logging levels. |
| src/test/kotlin/com/coder/toolbox/feed/IdeFeedManagerTest.kt | Updates logger setup and coroutine scope. |
| src/test/kotlin/com/coder/toolbox/feed/IdeFeedManagerOfflineTest.kt | Updates logger setup and coroutine scope. |
| src/test/kotlin/com/coder/toolbox/diagnostics/CoderSupportBundleCollectorTest.kt | Tests replay before support export. |
| src/test/kotlin/com/coder/toolbox/diagnostics/CoderLoggerTest.kt | Supplies logger settings. |
| src/test/kotlin/com/coder/toolbox/diagnostics/CoderLoggerFlightRecorderTest.kt | Tests capture, resizing, replay, and triggers. |
| src/test/kotlin/com/coder/toolbox/CoderRemoteProviderTest.kt | Supplies logger settings. |
| src/test/kotlin/com/coder/toolbox/CoderRemoteEnvironmentTest.kt | Supplies logger settings. |
| src/test/kotlin/com/coder/toolbox/cli/EnsureCLITest.kt | Uses a concrete coroutine scope. |
| src/main/resources/localization/defaultMessages.po | Adds buffer-setting interface text. |
| src/main/kotlin/com/coder/toolbox/views/NewEnvironmentPage.kt | Flushes on workspace-filter changes. |
| src/main/kotlin/com/coder/toolbox/views/CoderSettingsPage.kt | Exposes and validates buffer capacity. |
| src/main/kotlin/com/coder/toolbox/views/CoderPage.kt | Flushes after action failures. |
| src/main/kotlin/com/coder/toolbox/store/StoreKeys.kt | Adds the buffer-capacity persistence key. |
| src/main/kotlin/com/coder/toolbox/store/CoderSettingsStore.kt | Persists capacity and publishes settings changes. |
| src/main/kotlin/com/coder/toolbox/settings/ReadOnlyCoderSettings.kt | Defines observable settings and capacity limits. |
| src/main/kotlin/com/coder/toolbox/sdk/CoderRestClient.kt | Adds replay triggers for selected failures. |
| src/main/kotlin/com/coder/toolbox/diagnostics/FlightRecorder.kt | Implements ring buffering and batched replay. |
| src/main/kotlin/com/coder/toolbox/diagnostics/CoderSupportBundleCollector.kt | Flushes before collecting diagnostics. |
| src/main/kotlin/com/coder/toolbox/diagnostics/CoderLogger.kt | Captures string and lazy logs through the recorder. |
| src/main/kotlin/com/coder/toolbox/CoderToolboxExtension.kt | Constructs the shared settings-aware logger. |
| src/main/kotlin/com/coder/toolbox/CoderToolboxContext.kt | Reuses an existing Coder logger when provided. |
| src/main/kotlin/com/coder/toolbox/CoderRemoteProvider.kt | Adds provider lifecycle replay triggers. |
| src/main/kotlin/com/coder/toolbox/CoderRemoteEnvironment.kt | Adds environment lifecycle replay triggers. |
| README.md | Documents configuration, replay, diagnostics, and compatibility. |
| gradle/libs.versions.toml | Upgrades the Toolbox plugin API. |
| CHANGELOG.md | Records the feature and compatibility requirements. |
Review effort: Balanced
| textError ?: "", | ||
| ex, | ||
| ) | ||
| context.logger.flush("action_failed $description", sessionId) |
Collaborator
Author
There was a problem hiding this comment.
The flight recorder’s automatic error-triggered flushes are intended for failures that affect one or more SSH connections. This is consistent with the VS Code extension’s approach to connection-failure replay and maybe @EhabY and @aqandrew can back me up on this.
This connect screen can happen one of the 4 scenarios:
-
During fresh setup when there are no environments or SSH sessions yet, so an authentication or CLI setup failure cannot interrupt an existing connection.
-
After logout when the plugin cleans up its state, removes the existing environments, and flushes the recorder before presenting the setup wizard again.
-
When an expired API token requires signing in again, the same cleanup sequence (mentioned at 2) removes the existing environments and flushes diagnostics before returning to the wizard.
-
When a URI targets deployment 2 while the plugin is already connected to deployment 1. The existing deployment stays active while the new setup runs. A failure during deployment 2’s authentication or CLI setup does not itself terminate deployment 1’s SSH connections. Once the new setup succeeds, the plugin cleans up the previous deployment and flushes its diagnostics before installing the replacement.
I don’t think this handler needs an additional flush for the intended SSH-connection scope.
The SSH process search logs full command lines for every visited descendant and for the matching process. These arguments can contain credentials, and the flight recorder can persist them during automatic replay at INFO. This commit removes per-process command logging and limits the match diagnostic to the PID and workspace/agent name. A regression test covers token-bearing matching and unrelated processes and verifies that automatic disconnect replay retains the match diagnostic without exposing command arguments.
FlightRecorder captures timestamps before message preparation and insertion, so concurrent producers can replay entries with timestamps that disagree with their insertion order. This commit captures timestamps when prepared entries are accepted into the buffer, under the insertion lock. Message suppliers and exception formatting remain outside the lock so their execution does not block other producers. This keeps the timestamp semantics in sync with Coder's VS Code extension, where timestamps also mark acceptance of the prepared entry. The README documents acceptance order, and two-producer regression tests cover lazy evaluation, exception formatting, replay order, and capacity-one eviction.
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
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.
Up until today Toolbox discards plugin messages below its active log level, so connection failures can leave diagnostic logs without the preceding context unless verbose logging is enabled beforehand.
CoderLogger now captures filtered messages in a private flight recorder backed by a fixed-capacity ring buffer. Entries retain their original timestamp, severity, message, and exception details. Both string and lazy logging methods participate in recording, so future uses of lazy methods cannot silently bypass capture.
The
connectionLogBufferSizesetting defaults to 1000 entries, accepts values from 0 to 10000, and clears and disables recording at 0. Plugin settings expose changes through Kotlin flows, allowing the recorder to resize without a restart. Shrinking discards the oldest entries, while growing preserves existing entries.Connection lifecycle events and failures trigger replay into the Toolbox log files. Per-environment support bundle collection also flushes the recorder before export. Replay preserves chronological order and original timestamps and levels, groups up to 100 entries per call, and uses a severity accepted by the current Toolbox logging level. Replayed entries are removed from memory. Logging set to Off prevents both capture and replay.
An important note is that the Toolbox API upgrade to 1.15.88839 raises the minimum supported Toolbox version to 3.9 but the change is required in order to have access to the logging level configured in TBX.