Skip to content

Constants and timings

The fixed values in the code that shape the operator's behaviour, and the defaults of the flags that tune them. Each row names its source as file:line; line numbers can move between releases, so search for the constant name when a line does not match. The design behind these values is in the architecture.

Drift and requeue

value default source effect
DefaultDriftIntervalCritical 15 min internal/engine/drift.go:13 drift interval of the access-critical kinds listed in scale and drift; flag --drift-interval-critical (cmd/main.go:104)
DefaultDriftIntervalStandard 1 h internal/engine/drift.go:14 drift interval of every other kind; flag --drift-interval-standard (cmd/main.go:106)
MinDriftInterval 30 s internal/engine/drift.go:42 CEL rejects a shorter spec.driftInterval (api/common/v1alpha1/types.go:120); the engine raises one that bypassed validation to 30 s
mountTableTTL 2 s internal/controller/sys/mounttable.go:17 one read of sys/mounts or sys/auth serves every mount resource of the same connection and OpenBao namespace; a mount write or delete clears it, so an out-of-band mount change shows up to 2 s later
liveSnapshotTTL 10 s internal/engine/ownership.go:157 one live LIST of a kind serves the ownership rechecks before adoption; a path this process claimed after that LIST gets a fresh LIST
jitter × 0.8 to × 1.2 internal/engine/reconciler.go:1563-1564 applied to the drift requeue, to every backoff and to the pause recheck
minReferenceRetry 2 s internal/engine/reconciler.go:55 first delay of every backoff below
backoff growth time since Synced turned False with the same reason, at least 2 s internal/engine/reconciler.go:1554-1561 each retry waits about as long as the condition has lasted, so the delay roughly doubles
retryInterval 30 s internal/engine/reconciler.go:52 backoff cap for ReferenceNotReady and OutputNotDelivered
maxConnectionRetry 1 min internal/engine/reconciler.go:54 backoff cap for ConnectionNotReady; the connection watch re-queues sooner
OutOfSync cap the resource's drift interval internal/engine/reconciler.go:886 backoff cap after a write that did not read back as written
inProgressInterval 10 s internal/engine/reconciler.go:56 fixed requeue while OpenBao completes an asynchronous operation (ReconcileInProgress)
pausedRecheck 1 min internal/engine/reconciler.go:59 recheck of a resource paused through its Namespace or its OpenBaoServer; a pause on the resource itself waits for the annotation change

Request rate and recovery

value default source effect
driftShare 0.8 internal/openbao/ratelimit.go:14 requests without priority (drift reads, the first pass after an outage) use at most 80 % of OpenBaoServer.spec.rateLimit QPS and burst; spec changes may use all of it
rate limiter key server address internal/openbao/pool.go:72 one token bucket per address, shared by every connection to it; the most recent connection's rateLimit replaces the buckets
recoveryPerSecond 100 per second internal/engine/budget.go:23 rate of the first passes after ConnectionNotReady, per OpenBaoServer
recoveryBurst 10 internal/engine/budget.go:24 burst of the recovery ramp

Write budget

value default source effect
DefaultWriteBudgetPerMinute 60 internal/engine/budget.go:16, flag --write-budget-per-minute (cmd/main.go:111) rewrites per OpenBaoServer per window; -1 disables the cap
budgetWindow 1 min, sliding internal/engine/budget.go:18 window of the write budget
held-back retry the wait for the first free slot (at least 1 s) plus 0 to 60 s at random internal/engine/budget.go:96 spreads held-back rewrites over a window
budget state in process memory internal/engine/budget.go:29-42 a manager restart or a leader change starts with an empty window, so up to one full budget of rewrites per server can follow at once

Creates, adoptions, spec changes and changed input Secrets are not charged; the architecture describes which writes count.

OpenBao HTTP client

value default source effect
maxRetries 2 internal/openbao/client.go:22 retries per request after the first attempt, by the OpenBao API client's default retry policy; every attempt takes a rate-limit token
minRetryWait 250 ms internal/openbao/client.go:23 shortest wait between retries
maxRetryWait 2 s internal/openbao/client.go:24 longest wait between retries
redirects refused internal/openbao/client.go:274-280, :298-299 a 3xx answer is an error and is never followed, so a token or body is never re-sent to another host
maxAttributionLength 256 characters internal/openbao/ratelimit.go:24 longest X-Openbao-Config-Resource header; non-printable characters are dropped

Login, renewal and tokens

value default source effect
renewal point, renewable token a third of the TTL internal/openbao/pool.go:441 the token is renewed for its login TTL
renewal floor a third of the login TTL internal/openbao/pool.go:398 a renewal that returns a shorter lease (max TTL near) leads to a new login
replacement point, non-renewable token two thirds of the TTL internal/openbao/pool.go:443 a new login replaces the token
minReloginInterval 5 s internal/openbao/pool.go:21 after a 401, or a 403 whose auth/token/lookup-self fails, at most one new login per 5 s per client; inside that window the request fails without a login
sharedLoginTimeout 1 min internal/openbao/pool.go:28 limit of a login that several callers wait for
revokeTimeout 5 s internal/openbao/pool.go:24 limit of the best-effort revocation of a retired token
tokenExpirationSeconds 600 s internal/connection/resolver.go:57 lifetime requested for a TokenRequest token; each one is used for a single login

Server identity and probes

value default source effect
clusterIDTTL 1 min internal/openbao/clusterid.go:14 how long a client trusts the cluster id from sys/health; it is read again past the cache right before a create, a recreate and a delete
--cluster-id the UID of the kube-system Namespace cmd/main.go:117, :178-180, :260-266 written into ownership markers. A rebuilt Kubernetes cluster has a new kube-system UID, so markers written before the rebuild name another cluster and marked kinds (entities, groups, namespaces) report Conflict; a fixed --cluster-id survives the rebuild
probeInterval 1 min plus up to 20 % internal/controller/core/connection_controller.go:36, :189; internal/controller/core/server_controller.go:71 re-check of every OpenBaoServer and connection
namespace retry 2 s to 30 s internal/controller/core/connection_controller.go:45-46 retry while a connection's OpenBao namespace does not exist yet
inUseRecheck 5 s internal/controller/core/inuse.go:28 recheck of a connection or server held by its in-use finalizer

Hashing and redaction

value default source effect
hashKeySize 32 bytes internal/engine/hashkey.go:17 size of the key in the hash-key Secret (--hash-key-secret, default openbao-config-operator-hash-key, cmd/main.go:123)
minRedactLength 4 characters internal/engine/reconciler.go:1582 write-only string values shorter than this are not redacted from OpenBao error messages

Concurrency and leader election

flag default source effect
--max-concurrent-reconciles 4 cmd/main.go:108 workers per kind
--max-concurrent-reconciles-high-volume 8 cmd/main.go:109 workers of the kinds that usually have many objects (roles, policies, entities, groups)
--leader-elect true cmd/main.go:94 creates are serialised only within one process, so several replicas need it
GracefulShutdownTimeout 5 s cmd/main.go:164-165 after SIGTERM in-flight reconciles get 5 s to return, within the chart's 10 s terminationGracePeriodSeconds; the leader then releases its lease (LeaderElectionReleaseOnCancel), so the next pod takes over without waiting for the lease to expire