stzAppCluster

(no comment in the source: the name reads as a sentence) initno extension

declaring the fleet composition ------------------------------------ Declare nWorkers processes of a profile (capability tag + caps). One @aFleet row per worker is created NOW (port assigned at Start), so FleetSize/WorkersOf are correct before and after launch.QCSXTZZZIBWU

(no comment in the source: the name reads as a sentence) set max triesno extension

Circuit breaker: after nThreshold consecutive failures a worker's circuit OPENS (it is skipped for nCooldownMs, then half-open: one probe re-closes on success or re-opens on failure).no extension

Rate limit a facet at the front door: nRatePerSec sustained with a bucket of nBurst (the largest instantaneous burst absorbed). A facet with no SetRateLimit is UNLIMITED. Requests over the limit are shed with a -429 status before any worker is touched.no extension

Register a PRE-EXISTING worker (e.g. a remote/static host) into a facet's pool -- it participates in routing, failover and health like a spawned worker. host:port endpoint.no extension

THE GENERAL FORM: specialize workers along ANY facet in THIS cluster's catalog (not just the doc's four). Capabilities, provenance and any external tool are taken from the pool's own catalog (instance-scoped, customizable via CatalogQ()/DefineFacet). oC.WithFacet(:graph, 2).WithFacet(:knowledge, 2).WithFacet(:neural, 1)no extension

The cluster's facet catalog (instance-scoped -- customize per deployment: DefineFacet/DefinePolyglotFacet/Drop before WithFacet).no extension

Sugar for the doc's four (each just a WithFacet over the catalog).no extension

Same as WithFacet: THE GENERAL FORM: specialize workers along ANY facet in THIS cluster's catalog (not just the doc's four). Capabilities, provenance and any external tool are taken from the pool's own catalog (instance-scoped, customizable via CatalogQ()/DefineFacet). oC.WithFacet(:graph, 2).WithFacet(:knowledge, 2).WithFacet(:neural, 1)no extension

Same as WithFacet: THE GENERAL FORM: specialize workers along ANY facet in THIS cluster's catalog (not just the doc's four). Capabilities, provenance and any external tool are taken from the pool's own catalog (instance-scoped, customizable via CatalogQ()/DefineFacet). oC.WithFacet(:graph, 2).WithFacet(:knowledge, 2).WithFacet(:neural, 1)no extension

Same as WithFacet: THE GENERAL FORM: specialize workers along ANY facet in THIS cluster's catalog (not just the doc's four). Capabilities, provenance and any external tool are taken from the pool's own catalog (instance-scoped, customizable via CatalogQ()/DefineFacet). oC.WithFacet(:graph, 2).WithFacet(:knowledge, 2).WithFacet(:neural, 1)no extension

(no comment in the source: the name reads as a sentence) set worker ttlno extension

(no comment in the source: the name reads as a sentence) set base portno extension

(no comment in the source: the name reads as a sentence) set ring exeno extension

launching the fleetno extension

Poll every worker's /health until it answers 200 (workers need a moment to load stzBase + bind). Returns the number now ready.no extension

routing: RETRY-WITH-FAILOVER + per-worker CIRCUIT BREAKER ----------- Proxy to a healthy worker of pcTag. On a failed attempt (transport error or non-2xx), FAIL OVER to the next healthy worker, up to SetMaxTries. A worker that fails SetCircuitBreaker consecutive times has its circuit OPENED (skipped for the cooldown, then half-open). A success resets/closes its circuit. Path is SSRF/CRLF-validated first.no extension

(no comment in the source: the name reads as a sentence) route last statusno extension

(no comment in the source: the name reads as a sentence) whyno extension

observability (latency percentiles + trace ids)no extension

The trace id assigned to the most recent Route (correlates the front host's record with the worker's _trace on the wire).no extension

Tail-aware latency for a facet (ms; the engine histogram bucket bound).no extension

(no comment in the source: the name reads as a sentence) latency p90no extension

(no comment in the source: the name reads as a sentence) latency p99no extension

(no comment in the source: the name reads as a sentence) latency statsno extension

The last n request records: [ id, facet, endpoint, status, durMs, attempts ].no extension

rate limiting (front-door admission)no extension

How many requests this facet has shed for exceeding its rate limit.no extension

Tokens left in a facet's bucket (-1 = unlimited / no limit configured).no extension

(no comment in the source: the name reads as a sentence) circuit open countno extension

introspectionno extension

(no comment in the source: the name reads as a sentence) ready countno extension

(no comment in the source: the name reads as a sentence) workers ofno extension

(no comment in the source: the name reads as a sentence) portsno extension

(no comment in the source: the name reads as a sentence) pool qno extension

(no comment in the source: the name reads as a sentence) reactor qno extension

R8.2 smart routing (classify -> route) ----------------------------- A request classifier bound to THIS cluster's facet catalog (created lazily, after facets are declared).no extension

Classify a request to a facet (R8.2), then proxy it to a worker of that facet (R8.3). Returns the response body; "" (with negative RouteLastStatus) when the request is undecidable or has no worker.no extension

R8.4 health + elastic scale (the supervision surface) -------------- Re-probe every worker's /health and refresh its ready flag. A worker that stops answering (crashed, exited its TTL) flips to NOT ready. Returns the number ready now.no extension

Workers that were launched but are not answering (crashed / expired).no extension

Add ONE worker of a profile (elastic scale-up). Spawns a new process on a fresh port; call WaitReady()/HealthCheck() to see it come up. Returns the new port.no extension

GRACEFUL drain-down: mark one ready worker of a profile DRAINING -- routing stops immediately (it finishes in-flight work), and the process self-exits on its TTL. Returns the drained port, or 0 if there is nothing to drain. (Forced kill of a hung worker is a small future engine add; drain is the graceful path.)no extension

Respawn every dead worker in place (health-driven self-healing). Returns the number restarted.no extension

forced kill + orphan cleanup (the forceful sibling of drain) -------- Force-kill EVERY spawned worker of a facet (SIGKILL). The forceful counterpart to ScaleDown's graceful drain: use it on a WEDGED worker that neither answers health nor self-exits on its TTL. Marks each not ready and returns how many were actually killed. External workers (RegisterExternalWorker, jobId 0) are NOT ours to kill and are skipped.no extension

Total workers force-killed over the cluster's life (a hung-worker / orphan-cleanup signal for the health console).no extension

A per-profile metrics snapshot the supervisor reads (REAL counts).no extension

teardown ----------------------------------------------------------- R8.3: workers self-terminate on their TTL. Graceful drain / kill / health-restart / autoscale is R8.4 (stzAgentHost supervision).no extension