Repository navigation
feat: replace permanent event-delivery disable with a bounded cooldown #256
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Changes from all commits
828021b
0cd76e2
f40324a
8358410
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -319,6 +319,7 @@ static final class EventDispatcher { | |
| private final AtomicInteger busyFlushWorkersCount; | ||
| private final AtomicLong lastKnownPastTime = new AtomicLong(0); | ||
| private final AtomicBoolean disabled = new AtomicBoolean(false); | ||
| private final DeliveryCooldownTracker cooldown; | ||
| private final AtomicBoolean didSendInitEvent = new AtomicBoolean(false); | ||
| final DiagnosticStore diagnosticStore; // visible for testing | ||
| private final EventContextDeduplicator contextDeduplicator; | ||
|
|
@@ -346,6 +347,7 @@ private EventDispatcher( | |
| this.diagnosticStore = eventsConfig.diagnosticStore; | ||
| this.busyFlushWorkersCount = new AtomicInteger(0); | ||
| this.logger = logger; | ||
| this.cooldown = new DeliveryCooldownTracker(eventsConfig.cooldownNanoTimeFn); | ||
|
|
||
| ThreadFactory threadFactory = new ThreadFactory() { | ||
| @Override | ||
|
|
@@ -603,10 +605,21 @@ private boolean shouldDebugEvent(Event.FeatureRequest fe) { | |
| } | ||
|
|
||
| private void triggerFlush(EventBuffer outbox, BlockingQueue<FlushPayload> payloadQueue) { | ||
| if (disabled.get() || outbox.isEmpty()) { | ||
| if (disabled.get()) { | ||
| return; | ||
| } | ||
| // The cooldown check and the token issue are one atomic step, so a window opening | ||
| // concurrently on a worker thread either diverts this flush to the discard path or makes | ||
| // the dispatched payload's token stale. | ||
| DeliveryCooldownTracker.WindowToken token = cooldown.beginDeliveryUnlessInCooldown(); | ||
| if (token == null) { | ||
| discardPayloadInCooldown(outbox); | ||
| return; | ||
| } | ||
| if (outbox.isEmpty()) { | ||
| return; | ||
| } | ||
| FlushPayload payload = outbox.getPayload(); | ||
| FlushPayload payload = outbox.getPayload(token); | ||
| if (diagnosticStore != null) { | ||
| int summaryCount = 0; | ||
| for (EventSummary summary : payload.summaries) { | ||
|
|
@@ -632,12 +645,80 @@ private void triggerFlush(EventBuffer outbox, BlockingQueue<FlushPayload> payloa | |
| } | ||
| } | ||
|
|
||
| private void handleResponse(EventSender.Result result) { | ||
| // While a window is open, a flush of any kind is a no-op that behaves exactly as if the batch | ||
| // had been sent and the service had refused it: the buffered events are disposed of, and any | ||
| // context introductions they carried are forgotten so a later payload can introduce those | ||
| // contexts again. Only the network attempt and the serialization work are skipped. | ||
| private void discardPayloadInCooldown(EventBuffer outbox) { | ||
| if (outbox.isEmpty()) { | ||
| return; | ||
| } | ||
| // No batch reached the service, so the diagnostic batch-size stat must not report one. | ||
| if (diagnosticStore != null) { | ||
| diagnosticStore.recordEventsInBatch(0); | ||
| } | ||
| FlushPayload payload = outbox.getPayload(null); | ||
| outbox.clear(); | ||
| int eventCount = payload.events.length; | ||
| for (EventSummary summary : payload.summaries) { | ||
| if (!summary.isEmpty()) { | ||
| eventCount++; | ||
| } | ||
| } | ||
| boolean hadContextIntroductions = false; | ||
| for (Event e : payload.events) { | ||
| if (e instanceof Event.Index || e instanceof Event.Identify) { | ||
| hadContextIntroductions = true; | ||
| break; | ||
| } | ||
| } | ||
| if (hadContextIntroductions && contextDeduplicator != null) { | ||
| contextDeduplicator.flush(); | ||
| } | ||
| long remainingSeconds = Math.max(1, (cooldown.remainingMillis() + 999) / 1000); | ||
| if (cooldown.takeDiscardWarning()) { | ||
| logger.warn("Discarded {} events: event delivery is cooling down for another {} seconds", | ||
| eventCount, remainingSeconds); | ||
| } else { | ||
| logger.debug("Discarded {} events: event delivery is cooling down for another {} seconds", | ||
| eventCount, remainingSeconds); | ||
| } | ||
| } | ||
|
|
||
| private void handleResponse(EventSender.Result result, DeliveryCooldownTracker.WindowToken token) { | ||
| if (result.getTimeFromServer() != null) { | ||
| lastKnownPastTime.set(result.getTimeFromServer().getTime()); | ||
| } | ||
| if (result.isMustShutDown()) { | ||
| disabled.set(true); | ||
| if (result.isUnexpectedFailure()) { | ||
| noteUnexpectedFailure(token); | ||
| } else if (result.isSuccess()) { | ||
| if (cooldown.recordSuccess(token)) { | ||
| logger.info("Event delivery resumed after cooldown"); | ||
| } | ||
| } else { | ||
| cooldown.recordFailure(token); | ||
| } | ||
| } | ||
|
|
||
| // Diagnostic deliveries are not gated by the cooldown, and their outcomes drive it only the | ||
| // way they drove the old permanent disable: an unexpected failure opens a window, while a | ||
| // diagnostic success or ordinary failure has no effect on delivery state. In particular, a | ||
| // diagnostic success is not the probe and cannot count toward the reset threshold; only | ||
| // analytics deliveries prove the service is accepting analytics payloads again. | ||
| private void handleDiagnosticResponse(EventSender.Result result, DeliveryCooldownTracker.WindowToken token) { | ||
| if (result.getTimeFromServer() != null) { | ||
| lastKnownPastTime.set(result.getTimeFromServer().getTime()); | ||
| } | ||
| if (result.isUnexpectedFailure()) { | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. AI review: A diagnostic failure can open a window, but a diagnostic success can never end one. Suppose a proxy or allowlist accepts |
||
| noteUnexpectedFailure(token); | ||
| } | ||
| } | ||
|
|
||
| private void noteUnexpectedFailure(DeliveryCooldownTracker.WindowToken token) { | ||
| long windowMillis = cooldown.recordUnexpectedFailure(token); | ||
| if (windowMillis > 0) { | ||
| logger.error("Event delivery failed with an unexpected error;" | ||
| + " event delivery will not be attempted for the next {} seconds", windowMillis / 1000); | ||
| } | ||
| } | ||
|
|
||
|
|
@@ -650,9 +731,13 @@ public void run() { | |
| Writer writer = new BufferedWriter(new OutputStreamWriter(buffer, Charset.forName("UTF-8")), INITIAL_OUTPUT_BUFFER_SIZE); | ||
| gson.toJson(diagnosticEvent.value, writer); | ||
| writer.flush(); | ||
| // The send itself is not gated by the cooldown, but a diagnostic dispatched inside a | ||
| // window gets a null token, so its outcome cannot move delivery state even if it | ||
| // lands after the window has elapsed. | ||
| DeliveryCooldownTracker.WindowToken token = cooldown.beginDeliveryUnlessInCooldown(); | ||
| EventSender.Result result = eventsConfig.eventSender.sendDiagnosticEvent( | ||
| buffer.toByteArray(), eventsConfig.eventsUri); | ||
| handleResponse(result); | ||
| handleDiagnosticResponse(result, token); | ||
| if (diagnosticEvent.initEvent) { | ||
| didSendInitEvent.set(true); | ||
| } | ||
|
|
@@ -716,10 +801,10 @@ long getAndClearDroppedCount() { | |
| return res; | ||
| } | ||
|
|
||
| FlushPayload getPayload() { | ||
| FlushPayload getPayload(DeliveryCooldownTracker.WindowToken token) { | ||
| Event[] eventsOut = events.toArray(new Event[events.size()]); | ||
| List<EventSummarizer.EventSummary> summaries = summarizer.getSummariesAndReset(); | ||
| return new FlushPayload(eventsOut, summaries); | ||
| return new FlushPayload(eventsOut, summaries, token); | ||
| } | ||
|
|
||
| void clear() { | ||
|
|
@@ -731,15 +816,17 @@ void clear() { | |
| private static final class FlushPayload { | ||
| final Event[] events; | ||
| final List<EventSummary> summaries; | ||
| final DeliveryCooldownTracker.WindowToken token; | ||
|
|
||
| FlushPayload(Event[] events, List<EventSummary> summaries) { | ||
| FlushPayload(Event[] events, List<EventSummary> summaries, DeliveryCooldownTracker.WindowToken token) { | ||
| this.events = events; | ||
| this.summaries = summaries; | ||
| this.token = token; | ||
| } | ||
| } | ||
|
|
||
| private static interface EventResponseListener { | ||
| void handleResponse(EventSender.Result result); | ||
| void handleResponse(EventSender.Result result, DeliveryCooldownTracker.WindowToken token); | ||
| } | ||
|
|
||
| private static final class SendEventsTask implements Runnable { | ||
|
|
@@ -780,20 +867,33 @@ public void run() { | |
| } catch (InterruptedException e) { | ||
| continue; | ||
| } | ||
| EventSender.Result result; | ||
| try { | ||
| ByteArrayOutputStream buffer = new ByteArrayOutputStream(INITIAL_OUTPUT_BUFFER_SIZE); | ||
| Writer writer = new BufferedWriter(new OutputStreamWriter(buffer, Charset.forName("UTF-8")), INITIAL_OUTPUT_BUFFER_SIZE); | ||
| int outputEventCount = formatter.writeOutputEvents(payload.events, payload.summaries, writer); | ||
| writer.flush(); | ||
| EventSender.Result result = eventsConfig.eventSender.sendAnalyticsEvents( | ||
| result = eventsConfig.eventSender.sendAnalyticsEvents( | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. AI review: This sends without checking the cooldown. |
||
| buffer.toByteArray(), | ||
| outputEventCount, | ||
| eventsConfig.eventsUri | ||
| ); | ||
| responseListener.handleResponse(result); | ||
| } catch (Exception e) { | ||
| logger.error("Unexpected error in event processor: {}", LogValues.exceptionSummary(e)); | ||
| logger.debug(LogValues.exceptionTrace(e)); | ||
| // The batch was not delivered, and the delivery schedule must know that: an exception | ||
| // here counts as an ordinary failure, which breaks the consecutive-success streak | ||
| // without opening a cooldown window. | ||
| result = EventSender.Result.failure(); | ||
| } | ||
| try { | ||
| responseListener.handleResponse(result, payload.token); | ||
| } catch (Exception e) { | ||
| // An exception here (such as a null result from a custom EventSender) must not kill | ||
| // this worker thread: the busy-worker count would stay elevated forever and close() | ||
| // would never return. | ||
| logger.error("Unexpected error in event processor: {}", LogValues.exceptionSummary(e)); | ||
| logger.debug(LogValues.exceptionTrace(e)); | ||
|
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Null sender result skips failure recordingLow Severity A custom Additional Locations (1)Reviewed by Cursor Bugbot for commit 8358410. Configure here. |
||
| } | ||
| synchronized (activeFlushWorkersCount) { | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. AI review: Both catches above take only |
||
| activeFlushWorkersCount.decrementAndGet(); | ||
|
|
||
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -2,6 +2,7 @@ | |
|
|
||
| import com.launchdarkly.logging.LDLogger; | ||
| import com.launchdarkly.logging.LogValues; | ||
| import com.launchdarkly.sdk.internal.http.FailureClass; | ||
| import com.launchdarkly.sdk.internal.http.HttpHelpers; | ||
| import com.launchdarkly.sdk.internal.http.HttpProperties; | ||
|
|
||
|
|
@@ -16,7 +17,7 @@ | |
| import java.util.UUID; | ||
| import java.util.zip.GZIPOutputStream; | ||
|
|
||
| import static com.launchdarkly.sdk.internal.http.HttpErrors.checkIfErrorIsRecoverableAndLog; | ||
| import static com.launchdarkly.sdk.internal.http.HttpErrors.classifyHttpFailure; | ||
| import static com.launchdarkly.sdk.internal.http.HttpErrors.httpErrorDescription; | ||
|
|
||
| import okhttp3.Headers; | ||
|
|
@@ -152,7 +153,7 @@ private CompressionResult<byte[]> compressData(byte[] data) { | |
| private Result sendEventData(boolean isDiagnostic, byte[] data, int eventCount, URI eventsBaseUri) { | ||
| if (data == null || data.length == 0) { | ||
| // DefaultEventProcessor won't normally pass us an empty payload, but if it does, don't bother sending | ||
| return new Result(true, false, null); | ||
| return Result.success(null); | ||
| } | ||
|
|
||
| Headers.Builder headersBuilder = baseHeaders.newBuilder(); | ||
|
|
@@ -173,7 +174,7 @@ private Result sendEventData(boolean isDiagnostic, byte[] data, int eventCount, | |
| URI uri = HttpHelpers.concatenateUriPath(eventsBaseUri, path); | ||
| CompressionResult<byte[]> compressionResult = compressData(data); | ||
| RequestBody body = RequestBody.create(compressionResult.data, JSON_CONTENT_TYPE); | ||
| boolean mustShutDown = false; | ||
| boolean unexpected = false; | ||
|
|
||
| if (compressionResult.wasCompressed) { | ||
| headersBuilder.add("Content-Encoding", "gzip"); | ||
|
|
@@ -207,27 +208,24 @@ private Result sendEventData(boolean isDiagnostic, byte[] data, int eventCount, | |
| logger.debug("{} delivery took {} ms, response status {}", description, endTime - startTime, response.code()); | ||
|
|
||
| if (response.isSuccessful()) { | ||
| return new Result(true, false, parseResponseDate(response)); | ||
| return Result.success(parseResponseDate(response)); | ||
| } | ||
|
|
||
| String errorDesc = httpErrorDescription(response.code()); | ||
| boolean recoverable = checkIfErrorIsRecoverableAndLog( | ||
| logger, | ||
| errorDesc, | ||
| errorContext, | ||
| response.code(), | ||
| nextActionMessage | ||
| ); | ||
| if (!recoverable) { | ||
| mustShutDown = true; | ||
| if (classifyHttpFailure(response.code()) == FailureClass.UNEXPECTED) { | ||
| logger.error("Error {} (the payload will not be retried): {}", errorContext, errorDesc); | ||
| unexpected = true; | ||
| break; | ||
| } | ||
| logger.warn("Error {} ({}): {}", errorContext, nextActionMessage, errorDesc); | ||
| } catch (IOException e) { | ||
| checkIfErrorIsRecoverableAndLog(logger, e.toString(), errorContext, 0, nextActionMessage); | ||
| // Transport-level failures are always worth one retry here, whatever their cause; the | ||
| // request may simply have hit a transient network problem. | ||
| logger.warn("Error {} ({}): {}", errorContext, nextActionMessage, e.toString()); | ||
|
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Transport errors log unsanitized exceptionsMedium Severity The new IOException warn path logs Triggered by learned rule: Log exceptions with LogValues.exceptionSummary, not toString/getMessage Reviewed by Cursor Bugbot for commit 8358410. Configure here. |
||
| } | ||
| } | ||
|
|
||
| return new Result(false, mustShutDown, null); | ||
| return unexpected ? Result.unexpectedFailure() : Result.failure(); | ||
| } | ||
|
|
||
| private final Date parseResponseDate(Response response) { | ||
|
|
||


There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
AI review:
takeDiscardWarning()is atomic on its own, but not together with the cooldown check intriggerFlush, which takes the lock separately. A new window can open between them: the old window elapses, a diagnostic dispatched after that gets a 401, and the new window re-arms the Warn. This discard, gated by the old window, then takes the new window's Warn, and the new window's own discards log at Debug. In the Go PR, the check, the re-arm, and the check-and-set all run on the dispatcher goroutine, so this can't happen there. The impact is small, but returning the first-discard flag from the same tracker call as the cooldown check would close the gap.