[Luckydog691]

The hardcoded 1-hour lease expiration in nerdctl container commit can cause failures for containers with large filesystem deltas (e.g., many packages installed). When the diff operation exceeds 1 hour, the containerd GC cleans up the expired lease and its ingest data, resulting in context deadline exceeded / failed to export layer errors.

This PR continues the stalled #5108. The original commit by @Turbo-Jiaxxx789 is preserved as-is (full credit to the original author), rebased onto the latest main, with the outstanding review comment addressed in a follow-up commit.

Changes

• Add Timeout field to ContainerCommitOptions and commit.Opts
• Modify Commit() to use a configurable lease expiration:
  • --timeout=0 (or negative): use containerd's default lease expiration (24h), applied by client.WithLease when no lease opts are passed (see client/lease.go)
  • --timeout>0: use the specified duration
  • --timeout defaults to 1h for backward compatibility
• Add --timeout CLI flag (default: 1h)
• Add integration tests for --timeout=2h and --timeout=0s
• Add a unit test covering --timeout flag parsing

Addressing the remaining review comment from #5108

--timeout=0 does not parse (time.ParseDuration("0") needs a unit). Help/docs advertise 0, but cobra Duration only accepts 0s.

time.ParseDuration special-cases a bare "0" as a valid duration without a unit (go/src/time/format.go, "Special case: if all that is left is "0", this is zero."). Since pflag's Duration flag delegates to time.ParseDuration, both --timeout=0 and --timeout=0s parse successfully.

The added unit test TestCommitTimeoutFlagParsing locks in this behavior for 0, 0s, 90m, 4h, and the 1h default, and the command reference now explicitly documents that both 0 and 0s are accepted.

Usage

nerdctl container commit mycontainer myimage:tag      # default 1h
nerdctl container commit --timeout=4h ...             # 4h timeout
nerdctl container commit --timeout=0 ...              # 24h (containerd default)

Fixes #5107
Supersedes #5108

[Luckydog691]

The single failing check (in-host / nerdctl.gomodjail rootless) failed on TestRunRmTime (container_run_test.go:706: expected to have completed in 3s, took 12.9s), which is unrelated to this PR — it is the known flake tracked in #3988. The same job leg also fails on main at this PR's base commit (see the test workflow run for #5248). Going to close/reopen to retrigger CI.

[ChengyuZhu6]

Nit: the commit metadata looks inconsistent — could you please fix it ?

The commit author is hejianshan hejianshan@bytedance.com, but the Signed-off-by trailer is Turbo Jia turbo@example.com, which looks like a placeholder address.

[Luckydog691]

@ChengyuZhu6 Thanks for catching that. Fixed — the first commit keeps the original authorship (hejianshan <hejianshan@bytedance.com>) and the placeholder Signed-off-by: Turbo Jia <turbo@example.com> trailer has been corrected to Signed-off-by: hejianshan <hejianshan@bytedance.com>. No code changes (tree is identical).

[ChengyuZhu6]

LGTM