The ingest Job must return its sfdisk dump to the controller, and the container termination message is that Job's only outward channel. It is capped at 4 KiB.
cmd/ingest.boundResult already fails explicitly when the payload exceeds ingest.TerminationMessageLimit, naming the limit, rather than sending a truncated payload — so this fails loudly rather than corrupting an import.
Real GPT tables compress to well under 2 KiB, so this has never been hit. A disk with a pathological partition count would fail the import outright.
Alternatives, both larger changes:
- Read the pod log instead. Needs a clientset in the reconciler, which nothing there currently has.
- Have the ingest Job write a Kubernetes object itself. This breaks the rule that the ingest binary never talks to the API server.
Not scheduled. Filed so the limit is recorded somewhere other than a code comment.
The ingest Job must return its
sfdiskdump to the controller, and the container termination message is that Job's only outward channel. It is capped at 4 KiB.cmd/ingest.boundResultalready fails explicitly when the payload exceedsingest.TerminationMessageLimit, naming the limit, rather than sending a truncated payload — so this fails loudly rather than corrupting an import.Real GPT tables compress to well under 2 KiB, so this has never been hit. A disk with a pathological partition count would fail the import outright.
Alternatives, both larger changes:
Not scheduled. Filed so the limit is recorded somewhere other than a code comment.