v0.12.24

Surface the environment/arch header in redisbench-admin compare PR comments. Replaces the one-line 'Using platform named: X' with a structured block listing running_platform, triggering_env, baseline+comparison architectures, and baseline+comparison deployment types -- which matters now that PRs can produce both x86 and aarch64 benchmark rows in the same comment.

v0.12.23

Patch release: snap restart after parca-agent label set/clear.

Why

The parca-agent snap's `parca-agent-wrapper` reads `metadata-external-labels` via `snapctl get` once at service start and bakes the value into the `--metadata-external-labels=...` command-line flag. Runtime `snap set metadata-external-labels=...` (which 0.12.22 calls per-test) updates the stored snap config but leaves the running process untouched — so every sample shipped to Polar Signals with whatever value was baked in at boot (typically empty).

What

After every `snap set metadata-external-labels=...`, run `snap restart parca-agent.parca-agent-svc`. Sampling pauses for a few seconds while the eBPF profiler re-attaches — acceptable for multi-minute benchmarks.

• #529 — snap restart after label set/clear

v0.12.22

Patch release for per-test parca-agent label injection.

Changes since v0.12.21

• #525 — per-test parca-agent metadata-external-labels injection (test_name, git_hash, tested_commands, arch, topology, platform, coordinator_version, ...)
• run-remote: TESTING_INFRASTRUCTURE_REPO / TESTING_INFRASTRUCTURE_BRANCH env var overrides for on-the-fly terraform source
• docs(AGENTS.md): document tox as test runner + admin-merge caveat (#524)

When a DB node has parca-agent installed + active, run-remote will now push the benchmark metadata into the snap's metadata-external-labels before each test starts and clear them after, so every profile sample lands on Polar Signals tagged with the dimensions the query skills rely on.

v0.12.21

accept batch_size parameter and give ssh timeout a larger margin (#526)

v0.12.20

What's Changed

• Push benchmark results by module git_sha (with ft.debug auto-inference) by @fcostaoliveira in #520
• Push benchmark results keyed by architecture (x86 vs arm don't collide) by @fcostaoliveira in #521
• Bumping version from 0.12.19 to 0.12.20 by @fcostaoliveira in #522

Full Changelog: v0.12.19...v0.12.20

v0.12.19

improve error message and log summary (#519)

v0.12.18

What's Changed

• increase infra wait time by @JoanFM in #504

Full Changelog: v0.12.17...v0.12.18

v0.12.17

fix potential infinite hang (#518)

* fix potential infinite hang

* bump version

* fix: avoid infinite hang in execute_remote_commands

- Read stdout/stderr in background threads to prevent SSH transport
  window deadlock when output exceeds ~2MiB (paramiko/paramiko#448)
- Poll exit_status_ready() with wall-clock timeout instead of blocking
  on recv_exit_status() which ignores channel timeout (paramiko/paramiko#1815)
- Preserve original blocking behavior when no timeout is set

v0.12.16

fix standalone pass bigredis-max-ram (#517)

* fix standalone pass bigredis-max-ram

* bump version

v0.12.15

bump version (#516)