Ida is off until Monday, so this week's monitoring work is with you. The repo is /workspace/monitoring. Conventions: README, then docs/adr/. Ownership and SLO targets come from teams/ownership.yaml and slo/services.yaml, nowhere else. bin/range2test can replay any pull in incidents/ against your branch; CI is bin/af-check.
Queue
OPS-2684: storefront: one page per pod is too many
A database blip shouldn't page 14 times. cart-svc and returns-api lost their connections for a few minutes during the failover and every pod paged on its own; one page per service is what we want. incidents/INC-2968/.
OPS-2828: OrderEventsQueueBacklogHigh to page
We agreed in the ops review that OrderEventsQueueBacklogHigh should wake someone up. Right now it only posts to Slack.
OPS-2880: no page during the v65 crashloop
cart-svc v65 crashlooped for 39 minutes on Fri 11 Sep, about 39% of requests failed at the LB, and nobody was paged. Deploy log and thread: incidents/INC-2957/.
aino_str: linking the incident channel in case anyone wants the context
OPS-2690: drop NodeDiskFull
NodeDiskFull and HostDiskAlmostFull always fire together. Drop NodeDiskFull, it's the older of the two.
OPS-2600: 8 pages when eu-1 dropped
Reported by Olli Virtanen · 4 days ago · alerting
When eu-1 became unreachable last Tuesday, platform got ClusterUnreachable, which is right, and every service owner with pods there got paged as well, which is what README says shouldn't happen. order-events has moved into eu-1 since, so next time it would be one more. PD export in incidents/INC-2969/.
OPS-2707: OrderEventsLatencyHigh message is cut off
The Slack message for OrderEventsLatencyHigh just says "p99 latency above 500ms on" and then nothing. Which pod? Which service? I have to open Prometheus to find out.
Usual rules: everything lands in the repo, don't tidy what nobody asked about, and it's done when bin/af-check is green. promtool, amtool and bin/range2test are on the path.