Skip to content

Fortified IDDocs

Monitoring and resource limits

Health endpoints, periodic operational log lines, certificate lifetime metrics, the diagnostics command and request admission limits.

Signing Server reports its state through two health endpoints, a periodic operational log line and a local diagnostics command. Resource limits protect it from overload.

Endpoint Description
GET /health/live Process liveness.
GET /health/ready Checks key material, identity provider metadata validity and database availability without migrating.

Both return only UP or DOWN, with status 200 or 503. The listener limits are 128 open connections, 32 headers and 16 KiB of headers, 10 seconds to receive a request and 30 seconds to send a supplied/static DSS, CSC, seal or time stamp response. Dynamic human signing has a 180-second response ceiling. Configure client/proxy response timeouts for the selected host; no original authority, identity, issuance or consent deadline is extended.

After the first 30 seconds, and about every 30 seconds after that, the server writes one JSON line named signing.operations.v1 to the application log. It describes the process since it started, and counters restart at zero when the process restarts.

Member Content
operations Per stage, counts of SUCCESS, REJECTED, EXPIRED, CANCELLED, UNAVAILABLE and FAILED.
latency_bucket_counts Eight separate buckets with upper bounds of 1, 5, 20, 100, 500, 2000 and 10000 ms, and a last open bucket.
availability DATABASE, PROVIDER and READINESS as UNKNOWN, UP or DOWN with the age of the observation.
certificate_remaining_seconds Seconds until the earliest expiry in the chain for DOCUMENT, SAML_SIGNING, SAML_ENCRYPTION, DSS_ENVELOPE, TLS and ACCESS_SIGNING. Negative means expired, null means unknown.

The stages are HTTP_SIGN, HTTP_CONSENT, HTTP_CALLBACK, DSS_RESPONSE, AUTHORIZATION, DOCUMENT_CRYPTO, DATABASE, PROVIDER, READINESS, ADMISSION_WORK, ADMISSION_PENDING and ADMISSION_HEALTH. An HTTP success only means the response was below 400, and a signed error over HTTP 200 is counted as CANCELLED, EXPIRED or REJECTED, never as SUCCESS. Timing for a stage covers that stage only and is not end-to-end signing time.

The line contains no transaction identifiers, requester, signer or connection names, and no document, token or key data. Never use the counters as evidence of a transaction or of consent.

Organization seal and time stamp instances write signing.organization-seal.operations.v1 and signing.timestamp.operations.v1 with fixed HTTP status counts and a readiness flag.

Suggested actions:

Observation Action
READINESS is DOWN Run diagnostics and read the availability states. Never reset authorization to restore readiness.
PROVIDER is DOWN Check file permissions, the module hash, the token, the session and the PIN file.
DATABASE is DOWN Check the database, the network, TLS and the runtime account privileges.
No new line for a while Treat the state as unknown. Check the service manager, the log destination and the collector.
Certificate lifetime below your renewal window Replace that role through your approved procedure.
Rising FAILED or latency Investigate the database, provider and host load before you raise limits.
Terminal window
bin/signing-server diagnostics /protected/server.json

For the supplied/static DSS host, the command prints one signing.diagnostics.v1 JSON object. It reports the database and readiness as UP or DOWN, provider as NOT_PROBED and schema_unchanged as true. It starts no listener, loads no document key, signs nothing, uses no migration account and changes no state. Exit code 0 means both checks passed, 3 means unavailable or rejected and 2 means invalid usage.

With service="csc-v1", diagnostics validates the dedicated OAuth schema and runtime privileges. Its output identifies scope="oauth_schema_and_runtime_privileges_only", private_signing_probe=false and resumption_supported=false. It does not establish complete service readiness; use the running service health endpoint as well. CSC recovery-diagnostics reports CSC_RECOVERY_UNAVAILABLE and exits 3. Consumed operations are never resumed.

The DSS host always limits concurrent work and pending signing flows, and answers 503 with Retry-After: 1 when a limit is reached, without reading the body or consuming authority. An omitted resources object uses the defaults below. When you supply it, all five fields are required.

"resources": {
"untrusted_work": 4,
"trusted_work": 8,
"trusted_per_requester": 4,
"pending": 64,
"pending_per_requester": 16
}
Field Range Description
untrusted_work 1–32 Concurrent operations that are not part of an existing browser flow. Every first request to /sign counts here.
trusted_work 1–64 Concurrent operations that belong to an unexpired flow held by the server.
trusted_per_requester 1 to trusted_work One requester's share of the trusted work.
pending 1–256 Pending flows in total.
pending_per_requester 1–64, at most pending One requester's pending flows.

The two work budgets together must not exceed 96, at most 32 requesters can be registered, and the limit of 64 flows per connection also applies. A flow context only grants resources. It does not authenticate anyone, and the signed response, approval and every other check still apply. Admission never frees a consumed transaction: if a dependency fails after admission, do not replay the call.

Readiness probes use a single separate slot, and a probe at the same time receives DOWN. These limits do not replace limits in your proxy, and no service level is implied.