immich-app/yucca
1
0
Fork 0
mirror of https://github.com/immich-app/yucca.git synced 2026-09-30 13:33:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(ceph): let recovery outrank scrubbing on spice (#630)

Browse Source
This commit is contained in:
Andy Molenda
2026-09-08 12:29:30 -07:00
committed by GitHub
parent f0a5c49765
commit cadcba2389
1 changed files with 8 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tf/deployment/prod/htz-fsn1/ceph/clusters.auto.tfvars
+8 -12
View File
@@ -45,18 +45,11 @@ clusters = {
# its own copy for PG_NOT_DEEP_SCRUBBED, so an `osd` entry leaves the health
# check on the 7-day default.
#
# Two settings gate scrub scheduling. Both must be open or scrubbing stops.
#
# 1. Window off (0/0 = all day), overriding the role's 02:00-06:00. Tentacle
# Window off (0/0 = all day), overriding the role's 02:00-06:00. Tentacle
# has no `overdue` urgency, so the window gates a scrub however overdue it
# is. Client load here is hour-of-day independent, so the window protects
# nothing.
#
# 2. osd_scrub_during_recovery. The role default (false) does not scale with
# EC width: a recovering OSD refuses scrub reservations, and a k16+m4 PG
# needs all 20 shards free, so any nontrivial rebalance stops scrubbing
# outright. Costs roughly 4x recovery throughput while scrubs run.
#
# osd_scrub_cost and osd_deep_scrub_stride are upstream defaults. They and
# osd_scrub_disable_reservation_queuing (removed) came from the squid scrub
# thread: https://www.mail-archive.com/ceph-users@ceph.io/msg28007.html
@@ -65,13 +58,16 @@ clusters = {
osd_deep_scrub_interval = "2419200" # 28 days
}
osd = {
# Recovery IOPS outrank scrubbing. A k16+m4 PG needs all 20 shards free,
# so it goes unscrubbed for the length of any rebalance.
osd_scrub_during_recovery = "false"
osd_scrub_begin_hour = "0" # 0/0 = no window
osd_scrub_end_hour = "0"
osd_scrub_during_recovery = "true"
osd_max_scrubs = "6"
osd_scrub_cost = "52428800" # upstream default
osd_deep_scrub_stride = "524288" # upstream default, 512 KiB
osd_mclock_profile = "high_recovery_ops"
osd_scrub_cost = "52428800" # upstream default
osd_deep_scrub_stride = "524288" # upstream default, 512 KiB
osd_mclock_profile = "high_client_ops" # high_recovery_ops starves clients under backfill
osd_max_backfills = "6" # at 1, EC recovery reserves every peer and runs 2-3 PGs at a time
}
}
hosts = [