Files
Javier Aliaga 1a113f6f8e fix(perf): k6 settings for pubsub bulk and workflow scale perf tests (#10750)
* fix(perf): stop pubsub bulk perf test OOM-killing the k6 sidecar

The k6 publisher daprd is OOM-killed at the default 512Mi limit during
TestPubsubBulkPublishBulkSubscribeHttpPerformance for Pulsar. Raise its
limit to 1Gi, and raise the bulk p95 threshold for RabbitMQ and Pulsar
from 500ms to 2000ms to match current CI latency.

Signed-off-by: Javier Aliaga <javier@diagrid.io>

* fix(perf): raise k6 request timeout for delay_wf at 500 VUs

In TestDelayWorkflowsAtScale (t_500_10000) the slowest delay_wf
completions reach k6's 60s default request timeout, which fails requests
that are only slow. Use 300s for that scenario and keep 60s for the rest.

Signed-off-by: Javier Aliaga <javier@diagrid.io>

* fix(perf): raise RabbitMQ bulk subscribe threshold to 3000ms

The PR perf run measured a RabbitMQ bulk p95 of 2627ms with no failed
requests. Weekly runs ranged from 902ms to 1734ms. Use 3000ms, the same
as the RabbitMQ normal subscribe threshold.

Signed-off-by: Javier Aliaga <javier@diagrid.io>

---------

Signed-off-by: Javier Aliaga <javier@diagrid.io>
2026-10-13 13:49:27 +00:00
..
2026-02-25 08:44:13 -08:00
2026-02-25 08:44:13 -08:00
2026-02-25 08:44:13 -08:00
2026-02-25 08:44:13 -08:00
2026-05-12 12:15:41 -03:00
2023-12-01 01:02:39 +00:00