You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
iii-engine Docker container (iiidev/iii:0.11.2) fills disk space with massive 129GB stdout json.log #278
Describe the bug
When running the iii-engine via Docker (iiidev/iii:0.11.2) as part of the agentmemory stack, the container generates an extreme amount of stdout/stderr logs. In our environment, the container's /var/lib/docker/containers/.../...-json.log file reached 129 GB in size, completely consuming the host's disk space.
This appears to be caused by the engine getting stuck in a rapid crash/restart or error-spam loop, dumping massive amounts of stack traces or error logs to stdout.
To Reproduce
Run agentmemory Docker Compose stack with iiidev/iii:0.11.2
Allow it to run for some time (especially if it encounters an internal server error or network issue)
Check the container log file size (du -sh /var/lib/docker/containers/<container-id>)
Notice massive disk consumption.
Expected behavior
The iii-engine should not spam standard output infinitely on failure.
The docker-compose.yml for agentmemory should implement Docker logging limits (e.g., max-size: "10m", max-file: "3") to prevent unbounded disk growth.
Thanks @satabd — opened #280 with the compose-side guardrail you suggested: logging.driver: json-file with max-size: 10m and max-file: 3 on the iii-engine service. That caps total engine log on disk at 30 MB regardless of what the engine does inside.
The blast-radius bound is the right thing to land in agentmemory itself — but what's actually filling 129 GB is upstream iiidev/iii:0.11.2 getting stuck in a crash/restart or error-spam loop, and that needs fixing in iii-hq/iii, not here. To file that with concrete data instead of speculation, I need to see the spam pattern.
Can you share ~50-100 representative lines from your container log (sample from a few points across the 129 GB to catch repeating patterns vs one-shot stack traces)? Something like:
# tail the most recent activity
docker logs --tail 100 agentmemory-iii-engine-1 > /tmp/engine-tail.log
# sample from the middle of the giant log file (replace <id>)
CONTAINER_LOG="/var/lib/docker/containers/<id>/<id>-json.log"
head -n 100 "$CONTAINER_LOG"> /tmp/engine-head.log
tail -n 100 "$CONTAINER_LOG"> /tmp/engine-tail-raw.log
Specifically interested in:
Is it the same line repeating at high frequency, or many different errors?
Any stack traces with file paths into the engine binary?
A specific HTTP error / "internal server error" right before the spam starts?
Once I have that pattern I'll file a precise issue against iii-hq/iii. Until then #280 keeps your disk safe even when the engine misbehaves.
(Separately: if you can also share docker info | grep -i log and your Docker Desktop version, I can confirm whether the json-file driver default is what you're already on, since Docker Desktop on Windows occasionally defaults to a different driver where these compose options would be ignored.)
Activity
rohitg00 commentedon May 11, 2026
Thanks @satabd — opened #280 with the compose-side guardrail you suggested:
logging.driver: json-filewithmax-size: 10mandmax-file: 3on theiii-engineservice. That caps total engine log on disk at 30 MB regardless of what the engine does inside.The blast-radius bound is the right thing to land in
agentmemoryitself — but what's actually filling 129 GB is upstreamiiidev/iii:0.11.2getting stuck in a crash/restart or error-spam loop, and that needs fixing iniii-hq/iii, not here. To file that with concrete data instead of speculation, I need to see the spam pattern.Can you share ~50-100 representative lines from your container log (sample from a few points across the 129 GB to catch repeating patterns vs one-shot stack traces)? Something like:
Specifically interested in:
Once I have that pattern I'll file a precise issue against
iii-hq/iii. Until then #280 keeps your disk safe even when the engine misbehaves.(Separately: if you can also share
docker info | grep -i logand your Docker Desktop version, I can confirm whether thejson-filedriver default is what you're already on, since Docker Desktop on Windows occasionally defaults to a different driver where these compose options would be ignored.)fix(mcp): probe diagnostics + sandbox escape hatch (#234, #278) (#280)