Troubleshooting

Failures from bringing this Fargate lab up.

scmaster unhealthy, everything else looks dead

scmaster is the messaging bus. If ECS kills it during start, processors and the GUI lose 18180.

seiscomp update-config talks to localhost:18180 and can start a kernel scmaster before the run script execs the real process. Inventory and update-config trunk can run longer than the Fargate health startPeriod (capped at 300s). The container then fails health checks (often exit 137).

Logs: /ecs/seiscomp-ecs-lab stream ecs/scmaster/<task-id>. Wait until 18180 accepts TCP before treating other services as broken.

Client name not unique

Default SeisComP client name is the module name. Two scolv or two scautopick tasks collide. Processors set connection.clientName = $MODULE-$HOSTNAME. GUI sc-launch writes a per-process name in ~/.seiscomp. Do not start a second copy of the same GUI tool without that.

Processors empty, no bindings

Do not run seiscomp update-config in processor images. trunk.py always connects to localhost:18180. Bindings are written on the scmaster task.

Fargate tasks pending, vCPU

Eleven services at 0.25–0.5 vCPU add up. A simultaneous bounce can fail placement. deployment_maximum_percent = 100 is required for singletons (no two scmasters). Restart one service at a time.

var/run missing

seiscomp opens /home/sysop/seiscomp/var/run/seiscomp.pid. Entrypoints mkdir that path. A new image that forgets it dies immediately.

3389 times out

Your public IP changed. Re-apply with the current /32 as operator_cidr.

GUI still shows an old catalog

The GUI container does not reload MariaDB inventory by itself. Restart the GUI Docker container after scmaster has finished inventory, then open a new scrttv.