chore(metrics): remove the temporary seed-from-backup initContainer #49
Loading…
Reference in a new issue
No description provided.
Delete branch "chore/drop-seed-initcontainer"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Added on 2026-09-03 to reseed a VictoriaMetrics replica whose volume had been recreated on
hcloud-volumes. Its own comment said TEMPORARY — remove once both replicas are on hcloud-volumes and populated.Both conditions now hold:
victoriametrics-0victoriametrics-1Each holds the full 14-day window independently again.
Why not just leave it
It would make replica 0's object-storage backup a silent single point of truth. The container names
victoriametrics-0asSEED_SRCby literal, so any future empty replica would quietly restore from that one backup instead of failing loudly and letting a human decide. That is the wrong default for a recovery path.What stays
The
vmbackupsidecar, which is the actual durability story:s3://seanfarm/vm-backups/${POD_NAME}every six hours.vmrestorewas measured at 94 and 97 seconds for 19 GB during the migration, roughly 180x faster than thevmctl vm-nativecopy it replaced. Reseeding remains a one-command manual operation, which is the right shape for a deliberate decision.bash tests/check.shexits 0. The StatefulSet renders with no initContainers, both containers intact, and the claim template still onhcloud-volumes.https://claude.ai/code/session_01MdSbMhzabSbpG8TtP9Ur3H
It was added on 2026-09-03 to reseed a VictoriaMetrics replica whose volume had been recreated on hcloud-volumes, and its own comment said TEMPORARY: "Remove once both replicas are on hcloud-volumes and populated." Both conditions now hold. data-victoriametrics-0 and -1 are both on hcloud-volumes, and both answer a query at 13 days with 100 series, so each holds the full 14-day retention window independently again. Leaving it would make replica 0's object-storage backup a silent single point of truth: the container names victoriametrics-0 as SEED_SRC by literal, so any future empty replica would restore from that one backup rather than fail loudly and let a human choose. The vmbackup sidecar stays. It is the actual durability story, writing to s3://seanfarm/vm-backups/${POD_NAME} every six hours, and vmrestore was measured at 94 and 97 seconds for 19 GB. Reseeding remains a one-command manual operation, which is the right shape for something that should be a deliberate decision. Claude-Session: https://claude.ai/code/session_01MdSbMhzabSbpG8TtP9Ur3H