lukaszraczylo/kubemirror
1
0
Fork 0
mirror of https://github.com/lukaszraczylo/kubemirror.git synced 2026-10-31 10:09:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf(controller): stop per-target apiserver reads in mirror reconcile

Browse Source 
The mirror reconcile path issued one direct apiserver GET per target
namespace on every pass. With N sources and M target namespaces each
resync produced N*M GET requests. On a 65-namespace cluster this
generated 63 GET/s and overloaded one control-plane node.

Changes:
- reconcileMirror now trusts the informer cache. Content drift is
  detected via the source hash annotation. A stale resourceVersion
  makes Update return a conflict, which requeues the source.
- The verify-source-freshness flag now defaults to false. The Helm
  chart passes the value explicitly, so the chart default of false
  matches the binary behaviour.
- Reconcile performs one namespace LIST per pass. Target resolution
  and orphan cleanup share the result.
- Add load regression tests that count direct API reads in the
  reconcile hot path.
This commit is contained in:
lukaszraczylo
2026-08-29 13:44:54 +01:00
parent 8c0bb0c5b0
commit eaa1246fb7
6 changed files with 131 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cmd/kubemirror/main.go
+5 -5
View File
@@ -110,12 +110,12 @@ func main() {
"Burst limit for API server requests.")
flag.DurationVar(&resyncPeriod, "resync-period", 10*time.Minute,
"Period for resyncing all resources (catches updates missed due to informer cache delays).")
flag.BoolVar(&verifySourceFreshness, "verify-source-freshness", true,
flag.BoolVar(&verifySourceFreshness, "verify-source-freshness", false,
"Verify source resource freshness by comparing cache with direct API read. "+
"Prevents mirroring stale data and missed orphan cleanups when the informer "+
"cache lags behind watch events. Trade-off: one extra API call per reconcile "+
"when the cache is stale. Disable only if you are confident your cluster's "+
"watch latency is negligible.")
"Converges a few seconds faster when the informer cache lags behind watch "+
"events, at the cost of one direct apiserver GET per reconcile. Off by "+
"default: a stale read self-corrects on the next watch event, and update "+
"conflicts requeue automatically.")
flag.BoolVar(&lazyWatcherInit, "lazy-watcher-init", false,
"Enable lazy watcher initialization - only create informers for resource types that have resources marked for mirroring. "+
"Significantly reduces memory usage by avoiding watchers for unused resource types. "+