Skip to main content

Troubleshooting (symptom → cause)

Source: runbook 23. Measured in the 2026-07-13 incident (public 502 + container loop).

SymptomCauseWhat to do
Many Exited 128/255 at the same time, failed to bind host portprivate mesh flap with bind to mesh IP (mitigated 2026-07-14; if it returns, SOCIETY_BIND_IP=0.0.0.0 was lost)grep SOCIETY_BIND_IP /opt/sis/.env + mesh status; recreate with prod-up.sh
SIS port reachable from the public internetMissing DROP (new port since last sis-firewall.sh)sudo /opt/sis/tools/ops/sis-firewall.sh
502 on the domain; container “Up (healthy)”Orphan: Networks={}, Ports={} (came from containers start)Recreate with prod-up.sh
InvalidPasswordError / auth failed society@pgRecreated without --env-file or without prod overlayRecreate with prod-up.sh
Temporary failure in name resolution (base de datos relacional) in a loopNo container orchestrator networkRecreate with prod-up.sh
edge-proxy reload says config is unchanged and the vhost did not changeBind-mount by inode; mv/sed -iWrite by truncating the file; stat -c %i before and after
Let's Encrypt HTTP 429vhost added before DNS pointed at the boxWait; do not retry hot. Happened with chat.identia.do
Conversation UI (alt stack) will not start / no networkSiS is down or missing external network self-improving-society_defaultmake up (or prod) first; then cd infra/conversation-ui && containers compose up -d
Conversation UI (alt stack) 500 on /oauth/openidMissing DOMAIN_SERVER / DOMAIN_CLIENT in infra/conversation-ui/.envFill as in .env.example; restart sis-conversation-ui
502 on chat. while SiS is healthysis-conversation-ui stack down or bind not loopback after moving edge proxycontainers ps | grep sis-conversation-ui; conversation-ui_BIND=127.0.0.1 if edge proxy is local
503 on /ontology-catalog/linkMissing SOCIETY_ontology-catalog_* in the box compose, not (only) in .envanadir_ontology-catalog_compose.py + recreate telemetry
Ontology button disabledMissing SOCIETY_ontology-catalog_PUBLIC_URLSee the button title
Ontology tab blank, HTTP 200Explorer served under a Console pathIt must be onto.<client>.do (own origin)
containers compose shows ontology catalog ports: and nothing listensinternal: true network ignores publishThat is the relay (ontology-catalog-rele :8010)
messaging channel mute after restartPID 1 does not re-read sticky; or creds.json was lostselect the profile in the harness CLI + restart; session is not reproducible
Message to the bot with no reply, same phone that scanned the QRYou are messaging yourselfTest from another account
ci-check green / red depending on the shellvenv in front or notUse ci-check / uv run from the root, not bare pytest
Deliverable signed by “the wrong writer”Resolve-by-name onlyAlready patched: identity = pair. If old data has no author, reconciliation does not rewrite

Recreating relational database does not wipe pg-data. A backup.sh restore does.