Replace the 3s $interval polling of container logs with a live HTTP
stream, and stop re-writing already-rendered lines (fixes selection bug).
- streamContainerLogs (containers.service.ts): fetch + ReadableStream
reader with follow=1, same-origin credentials:'include' (httpOnly JWT
cookie; CSRF only guards mutations), agent-target / manager-operation
headers replicated for Agent/Edge, AbortSignal-driven lifetime.
- containerLogsController: stream instead of poll; append parsed lines
into the buffer (push, never replace), cap at 5000 lines trimming from
the head; AbortController on pause/destroy/param-change; reconnect with
3s backoff resuming from `since` (dropping tail) on stream end/error;
Live toggle pauses/resumes the stream; tail/since/timestamps changes
restart the stream.
- log-viewer: `track by log.id` (was $index), filtering moved out of the
template into the controller (applyFilter via $watchCollection), removed
inert force-glue, decoupled auto-scroll from log collection, relabelled
"Auto-refresh logs" -> "Live logs", clearer empty states.
Backend unchanged (logs already stream transparently through the Docker
proxy). Shared task/service log views keep working via the new id'd lines.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
claude code agent
c3cdb8007e