PMM-15398 Build dashboards memory fix

[theTibi]

Feature build for PMM-15398 — node Used memory now counts PostgreSQL shared_buffers.

Used was MemTotal - (MemFree + Buffers + Cached), and /proc/meminfo:Cached includes Shmem, so shared_buffers was subtracted out of Used and re-drawn as Cache. This build carries the corrected expressions so QA can check the OS dashboards against free on a real PostgreSQL node.

Component PRs

• percona/pmm — PMM-15398 Count shared memory as used, not cache percona/pmm#5855

ci.yml

deps:
  - name: pmm
    branch: PMM-15398-memory-used-shmem

What to look at

Open OS → Node Summary → Memory Utilization for a PostgreSQL node with a large shared_buffers and compare Used against free -m on that host. Also check Memory Details → Memory Usage, Nodes Compare, and Nodes Overview → Top 5 Used Memory.

Two things worth checking explicitly:

• The stacked panel must still sum to MemTotal — Used + Free + Buffers + Cache.
• An RDS node must be unchanged, not blank. rds_exporter does not export SReclaimable or Shmem, and the expressions carry a zero-fallback for exactly that case.

Used reads higher than before on every node — that is the fix, not a regression.

Ticket

• https://perconadev.atlassian.net/browse/PMM-15398

[JNKPercona]

Server docker: perconalab/pmm-server-fb:PR-4550-77f19d7
Client docker: perconalab/pmm-client-fb:PR-4550-77f19d7
Client tarball: https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4550-77f19d7.tar.gz
Staging instance: https://pmm.cd.percona.com/job/pmm3-aws-staging-start/parambuild/?DOCKER_VERSION=perconalab/pmm-server-fb:PR-4550-77f19d7&CLIENT_VERSION=https://s3.us-east-2.amazonaws.com/pmm-build-cache/PR-BUILDS/pmm-client/pmm-client-PR-4550-77f19d7.tar.gz

[JNKPercona]

API tests have succeded: https://pmm.cd.percona.com/job/pmm3-api-tests/7192/