reposerver.parallelism.limit |
argocd-cmd-params-cm |
0 (Unbounded) |
5 (CMP v2) | 6 (Helm Umbrella) | 8 (Kustomize) | 14 (Plain YAML/OCI) |
Caps concurrent helm template and kustomize build subprocesses per replica so webhook cache flushes wait on a Go semaphore instead of triggering OOMKilled 137. |
reposerver.exec.timeout |
argocd-cmd-params-cm |
90s |
90s (OCI) | 120s (Kustomize) | 150s (Helm) | 180s (CMP Sidecar) |
Prevents queued chart renders from aborting with Context Deadline Exceeded while waiting behind the parallelism semaphore. |
controller.status.processors |
argocd-cmd-params-cm |
20 |
clamp(ceil(apps / 10) * 5, 20, 160) |
Scales concurrent Kubernetes live-state diff workers across Application reconciliations to prevent multi-minute sync status lag. |
controller.operation.processors |
argocd-cmd-params-cm |
10 |
clamp(ceil(status.processors / 2), 10, 80) |
Controls concurrent kubectl/Server-Side Apply sync operations executing against target spoke clusters. |
controller.sharding.algorithm |
argocd-cmd-params-cm |
legacy |
"round-robin" when clusters >= 4 else "legacy" |
Distributes managed spoke cluster informer watches evenly across argocd-application-controller StatefulSet ordinals. |
timeout.reconciliation (+ .jitter) |
argocd-cm |
180s (0s jitter) |
180s..600s + 20% reconciliation jitter |
Extends background Git polling intervals once webhooks are active and desynchronizes reconciliation spikes across hundreds of Applications. |
resource.exclusions |
argocd-cm |
Minimal |
Excludes Event (events.k8s.io) & CiliumIdentity churn |
Cuts application-controller Go heap consumption by 30%–45% on busy multi-tenant clusters without affecting workload health checks. |