Straza / snapshot distribution

Policy snapshot distribution timeline Activating a PolicySet compiles and signs a content-addressed snapshot; live clients adopt it by pull over three lanes: a busy session through the post-decision check of straza service within about 30 seconds, an idle session through its next call or renewal, and a new session at checkin. A fail-closed grace boundary denies everything governed if no valid signed snapshot is available. activate admin API · strazactl compile < 2 s @ 10k rules sign ed25519 snapshot id 9f3a…c1 · CBOR store + serve GET /v1/snapshot · ETag CLIENT ADOPTION: PULL, BUSY SESSION ≤ 30 s t=0 10 s 20 s 30 s ≤ 30 s t=0.0 s busy session post-decision checkcheck snap 41d2…8csnap 9f3a…c1 next decision after the throttle window idle session next call or renewalnext call snap 41d2…8csnap 9f3a…c1 its next call new session checkin · token refreshcheckin (no snapshot)snap 9f3a…c1 fetched at session start fail-closed boundary security layer unreachable + no valid signed snapshot within the grace TTL ⇒ deny everything governed grace TTL: enterprise 0 · standalone 15 min adoption is pull-based plus a push nudge: sessions with a push stream adopt in < 5 s, the next call is the fallback; the < 2 s lever is the session kill switch
Press “activate a policy” to compile, sign and watch the three adoption lanes pick up the new snapshot. Hover any element for its facts.
What this shows

Activating a PolicySet compiles every active set into one snapshot (< 2 s at 10k rules), signs it with the ed25519 snapshot key, and stores it; the snapshot id is the content address of the CBOR blob, served over GET /v1/snapshot with an ETag. Distribution is pull-based, with a push nudge for idle sessions: a busy session adopts it via the post-decision check of straza service (a background conditional GET, at most once per the managed config's snapshotLagSeconds for each user, default 30 s; a busy session lags at most one tool call plus that); an idle session whose push stream is up gets the policy-update push nudge and adopts in under 5 s, and without one no timer runs: its first call after a change still decides on the old snapshot unless that call renews the session, and the check it starts brings the new one for the calls after it; new sessions and token refreshes fetch it at checkin. In every lane the blob is fetched and its signature verified before the session's snapshot id advances. An id alone would brick hooks. If the security layer is unreachable and no valid signed snapshot exists within the grace TTL (enterprise default 0, standalone 15 min), the client fails closed and denies everything governed. When a change must land in under 2 seconds, the lever is the session kill switch, not policy distribution.