PMM-14665 skip-non-eligible-nodes-from-add-service - #4498

Server docker: perconalab/pmm-server-fb:PR-4498-985e3eb
Client docker: perconalab/pmm-client-fb:PR-4498-985e3eb
Watchtower docker: perconalab/pmm-watchtower-fb:PR-4498-985e3eb
Client tarball: https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4498-985e3eb.tar.gz
Staging instance: https://pmm.cd.percona.com/job/pmm3-aws-staging-start/parambuild/?DOCKER_VERSION=perconalab/pmm-server-fb:PR-4498-985e3eb&CLIENT_VERSION=https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4498-985e3eb.tar.gz

API tests have succeded: https://pmm.cd.percona.com/job/pmm3-api-tests/6891/

Server docker: perconalab/pmm-server-fb:PR-4498-a52910e
Client docker: perconalab/pmm-client-fb:PR-4498-a52910e
Watchtower docker: perconalab/pmm-watchtower-fb:PR-4498-a52910e
Client tarball: https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4498-a52910e.tar.gz
Staging instance: https://pmm.cd.percona.com/job/pmm3-aws-staging-start/parambuild/?DOCKER_VERSION=perconalab/pmm-server-fb:PR-4498-a52910e&CLIENT_VERSION=https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4498-a52910e.tar.gz

API tests have succeded: https://pmm.cd.percona.com/job/pmm3-api-tests/6896/

Server docker: perconalab/pmm-server-fb:PR-4498-700bc38
Client docker: perconalab/pmm-client-fb:PR-4498-700bc38
Client tarball: https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4498-700bc38.tar.gz
Staging instance: https://pmm.cd.percona.com/job/pmm3-aws-staging-start/parambuild/?DOCKER_VERSION=perconalab/pmm-server-fb:PR-4498-700bc38&CLIENT_VERSION=https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4498-700bc38.tar.gz

API tests have succeded: https://pmm.cd.percona.com/job/pmm3-api-tests/7186/

Both red checks on run 33155754795 were investigated and neither is caused by your change — I re-ran the two failed jobs and both came back green (@docker-configuration, @fb-instances), so your PR is clear.

@docker-configuration did reproduce on a throwaway VM: the external-ClickHouse BeforeSuite runs docker compose up -d, and compose aborts with container pmm-server-external-clickhouse is unhealthy because the PMM Server image flags itself unhealthy ~37s after start (start-period 25s + 3×4s) while a cold start needs 45-50s to answer readyz. That is the product-side race already tracked as PMM-15361 (In Review); the QA-side fix so our test stops losing it is percona/pmm-qa#1270.

@fb-instances failed earlier, in setup: the HAProxy container could not publish host port 42100 (address already in use). 42100 sits inside the runner's ephemeral port range (32768-60999), so a concurrent outbound socket can transiently hold it while the three --parallel setups pull images — a transient collision, and it did not reproduce. PMM-QA is tracking that one for hardening.

Generated by Claude Code