@@ -51,10 +51,11 @@ the hub is [`../README.md`](../README.md).
5151 show what you have; an
5252 older Compose rejects the compose file with an error on ` required: ` before
5353 anything starts.
54- - 12 GiB RAM recommended and 20 GiB free disk. With an orchestrator release that
55- reads ` PAUSE_ADMISSION_DISK_HEADROOM_MIB ` (the pin here predates it), a pause
56- the disk cannot hold is refused before it starts, keeping 1 GiB free, instead
57- of failing midway. ` HUGEPAGES=2048 ` reserves
54+ - 12 GiB RAM recommended and 20 GiB free disk. A pause the disk cannot hold is
55+ refused before it starts, keeping 1 GiB free
56+ (` PAUSE_ADMISSION_DISK_HEADROOM_MIB ` ), instead of failing midway; the
57+ estimate counts the sandbox's whole memory, so a large sandbox needs that
58+ much free disk to pause. ` HUGEPAGES=2048 ` reserves
5859 4 GiB of that RAM for sandboxes, and ` preflight ` does not check RAM, so on
5960 a smaller host the first signal is ` host-setup ` failing with `FIX: give the
6061 host more memory (12 GiB recommended) or lower HUGEPAGES`. Lower
@@ -391,12 +392,9 @@ restarts; a single machine has nowhere to drain them to.
391392- The team API key rotates only through the seed (set ` TEAM_API_KEY ` , or
392393 remove the key file, then ` up ` ), not through an API call, and the old key
393394 keeps working for up to five minutes afterwards.
394- - With an api release that reads ` PAUSE_REFUSAL_RESTORE ` (the pin here
395- predates it), a pause the node refuses — it is still persisting the
396- sandbox's parent — leaves the sandbox running and answers 503; retry it.
397- Until then a refused pause ends the old way, with the sandbox killed and a
398- 500 . A full disk is not a refusal in the pinned orchestrator: that pause
399- fails after it has started and ends the same old way.
395+ - A pause the node refuses — it is still persisting the sandbox's parent, or
396+ the disk cannot hold the capture — leaves the sandbox running and answers
397+ 503 (` PAUSE_REFUSAL_RESTORE ` ); retry it.
400398- Container logs are capped: every service keeps at most five 50 MiB log
401399 files (` x-logging ` in ` compose.yaml ` ), so ` docker compose logs ` shows about
402400 the last 250 MiB per service and older lines are gone. Without the cap a
0 commit comments