libbeat: report queue filled.pct as 0 when the queue has no finite limit - #54652

Proposed commit message

The disk queue treats max_size: 0 as unlimited: it reports MaxBytes(0) to the queue observer and never sets a maximum event count. queueObserver.updateFilledPct() in libbeat/publisher/queue/monitoring.go guarded the byte division with maxBytes > 0 but then divided by the event maximum unconditionally, so queue.filled.pct became +Inf while events were queued and NaN when the queue was empty. The non-finite value made the monitoring reporter fail with "json: unsupported value: +Inf" and broke the /stats endpoint, as reported in #54639.

Fix: only compute the percentage when a finite byte or event limit exists. When the queue is unbounded, report 0. The filled.pct gauge is initialized to 0, and an unbounded queue can never fill up. The finite-limit paths are unchanged.

Tests: libbeat/publisher/queue/monitoring_test.go is new (the package previously had no test files). Its two unlimited-queue cases fail on the old code with exactly +Inf and NaN, and pass with the fix. A third case pins the finite byte and event percentages (0.25 and 0.5) so the existing behavior cannot drift. go test ./libbeat/publisher/queue/... passes (queue, diskqueue, memqueue, slabqueue).

A changelog fragment (kind: bug-fix, component libbeat) is included in this PR.

Checklist

• My code follows the style guidelines of this project
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation (no documentation describes this metric's behavior for unlimited queues; the fix restores the documented /stats output)
• I have made corresponding change to the default configuration files (no configuration change; max_size: 0 keeps its existing meaning of unlimited)
• I have added tests that prove my fix is effective or that my feature works. Where relevant, I have used the stresstest.sh script to run them under stress conditions and race detector to verify their stability. (Unit tests only; stress testing is not relevant to this observer calculation.)
• I have added an entry in ./changelog/fragments using the changelog tool.

Disruptive User Impact

None. The only change is the value of queue.filled.pct for queues configured without a finite limit, where it previously held +Inf or NaN and broke monitoring output.

How to test this PR locally

go test ./libbeat/publisher/queue/...

To reproduce the original failure, run the two unlimited-queue tests in libbeat/publisher/queue/monitoring_test.go against the code without the monitoring.go change: they fail with +Inf and NaN. End to end: run Filebeat with a disk queue configured with max_size: 0, publish events, and request /stats; before the fix the monitoring reporter logs "json: unsupported value: +Inf", after the fix queue.filled.pct reports 0.

Related issues

• Closes Filebeat disk queue with max_size: 0 causes monitoring metrics and /stats to fail #54639

Use cases

Users running Filebeat with an unlimited disk queue (max_size: 0, a valid configuration) get a working /stats endpoint and monitoring output again; queue.filled.pct reports 0 instead of a non-finite value.

Logs

New tests against the old code:

--- FAIL: TestQueueObserverFilledPctUnlimitedQueue — filled.pct must not be infinite, got +Inf
--- FAIL: TestQueueObserverFilledPctUnlimitedEmptyQueue — got NaN
--- PASS: TestQueueObserverFilledPctFiniteLimits

With the fix, all three tests pass and go test ./libbeat/publisher/queue/... is green across queue, diskqueue, memqueue, and slabqueue.