Configurable wait timeout for process to end and `-w SECONDS` CLI option by iandennismiller · Pull Request #363 · rfjakob/earlyoom

iandennismiller · 2026-02-17T14:56:17Z

This pull request introduces a new configurable option, -w SECONDS, to control the maximum time earlyoom waits for a killed process to exit. The implementation includes updates to the CLI, input validation, documentation, and associated tests. This change allows users to adjust the wait timeout (default 10 seconds) to better handle processes that take longer to terminate.

New feature: Configurable process wait timeout

Added a -w SECONDS command-line option to set the maximum wait time (in seconds) for a killed process to exit, with input validation and a maximum limit of 24 hours (86400 seconds). [1] [2] [3]
Updated the poll_loop_args_t struct and related logic to store and use the new kill_wait_timeout_secs value, defaulting to 10 seconds. [1] [2]
Modified the process waiting logic in kill_wait to use the user-specified timeout, with overflow protection for the poll loop. [1] [2]

Documentation and testing

Updated the documentation (MANPAGE.md) to describe the new -w option and its usage.
Added and updated CLI tests to verify correct handling of the -w flag, including invalid values and edge cases.

Internal constants and options

Introduced a new constant for the maximum allowed wait timeout and updated the CLI options enum accordingly. [1] [2]

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

…afety Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

…wait-countdown Add configurable kill_wait() timeout via -w flag

rfjakob · 2026-02-17T19:31:53Z

Hi, what problem does this solve?

iandennismiller · 2026-02-17T20:35:22Z

The process I'm monitoring takes longer than 10 seconds to shut down; and if it is terminated forcefully, it results in database corruption.

I need to prevent earlyoom from re-issuing SIGTERM every 10 seconds. Eventually, sending enough SIGTERMs will cause the process to terminate early, which corrupts the database. So, after 100 seconds and the receipt of 10 SIGTERMs, it will exit immediately as if it were SIGKILLed.

With the flag in this PR, I am specifying -w 300 to to throttle how quickly earlyoom sends SIGTERM. The desired outcome is that subsequent SIGTERMs are sent less frequently. In practice, I haven't seen the process take longer than 900 seconds to finish. So for me, a wait time of 5 minutes fits the time scale of my monitored process better than the 10 second interval.

rfjakob · 2026-02-17T20:36:52Z

So your process counts SIGTERMs? What process is this?

iandennismiller · 2026-02-17T20:37:47Z

go-ethereum

rfjakob · 2026-02-17T20:44:02Z

Wow, really https://github.com/ethereum/go-ethereum/blob/3eed0580d4dc40fb15a14e1b0e9fc6985fefbfd7/cmd/utils/cmd.go#L110

iandennismiller · 2026-02-17T20:46:01Z

Wow, really https://github.com/ethereum/go-ethereum/blob/3eed0580d4dc40fb15a14e1b0e9fc6985fefbfd7/cmd/utils/cmd.go#L110

Exactly; a bit idiosyncratic but there are non-production situations where it's handy, I suppose.

But for my purposes, I need to work around the 10 SIGTERMS behavior because it's not what I intend.

rfjakob

Some small changes, otherwise ok.

rfjakob · 2026-02-17T21:03:53Z

main.c


    int c;
-    const char* short_opt = "m:s:M:S:kingN:P:dvr:ph";
+    const char* short_opt = "m:s:M:S:kingN:P:dvr:phw:";


Please make this --kill-wait instead of -w.

rfjakob · 2026-02-17T21:05:21Z

main.c

                "                            to 0 to disable completely\n"
                "  -p                        set niceness of earlyoom to -20 and oom_score_adj to\n"
                "                            -100\n"
+                "  -w SECONDS                max seconds to wait for a process to die (default 10)\n"


--kill-wait

rfjakob · 2026-02-17T21:05:29Z

MANPAGE.md

    OOMScoreAdjust=-100
    Nice=-20

+#### -w SECONDS


--kill-wait

rfjakob · 2026-02-17T21:05:57Z

kill.c

+    if ((unsigned)args->kill_wait_timeout_secs > UINT_MAX / 10u) {
+        // Cap at maximum to avoid overflow (should never happen in practice)
+        max_poll_iterations = UINT_MAX;


Cannot happen. Delete the check.

rfjakob

Please rebase to master

Currently fails with Minimum supported API version is 1.44, please upgrade your client to a newer version. addnab/docker-run-action#62

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 2 to 4. - [Release notes](https://github.com/github/codeql-action/releases) - [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md) - [Commits](github/codeql-action@v2...v4) --- updated-dependencies: - dependency-name: github/codeql-action dependency-version: '4' dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com>

Bumps [actions/checkout](https://github.com/actions/checkout) from 2 to 6. - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](actions/checkout@v2...v6) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: '6' dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com>

As mentioned in issue rfjakob/earlyoom#358 and README, after Linux 5.17, OOM sysrq will always kill a process. This implements an option for trigger the kernel oom killer i.e. write an `f` to `/proc/sysrq-trigger`. Since the victim cannot be chosen by earlyoom, some of the features about customized victim selection won't take effect.

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

iandennismiller · 2026-02-18T15:39:06Z

I am going to change the sequencing of this pull request and re-submit a new one in a bit. The problem is that I merged to master and based the PR on that, when I should have based the PR on the pre-merge branch. It will be simpler for me to cherry-pick the changes into a new branch that's already synced to yours.

iandennismiller · 2026-02-18T16:37:12Z

Closing this PR.

Please see #366 for the replacement PR with a cleaner merge path.

Copilot AI and others added 10 commits February 17, 2026 14:16

Initial plan

d529256

Implement parameterizable kill_wait timeout with -w flag

b543921

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Address code review feedback: improve input validation and overflow s…

198d287

…afety Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Final refinements: add MAX constant and improve type casting

7b276d1

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Security check complete - no vulnerabilities found

7a7ec64

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Delete _codeql_detected_source_root

be764b1

revert superficial change

d67149d

Rename --kill-wait-timeout to --wait for brevity

92160bb

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Remove --wait long argument, keep only -w short form

0a6f7cb

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Merge pull request #1 from iandennismiller/copilot/parameterize-kill-…

3db01b9

…wait-countdown Add configurable kill_wait() timeout via -w flag

rfjakob requested changes Feb 17, 2026

View reviewed changes

iandennismiller marked this pull request as draft February 17, 2026 21:59

rfjakob and others added 10 commits February 18, 2026 10:20

Configure Dependabot for GitHub Actions updates

789d738

tests: set default comm to "foo"

a6d1afe

ci: update docker-run-action

ebe8d53

Currently fails with Minimum supported API version is 1.44, please upgrade your client to a newer version. addnab/docker-run-action#62

Initial plan

de77ec8

Implement parameterizable kill_wait timeout with -w flag

acfd250

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Rename --kill-wait-timeout to --wait for brevity

88baf04

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

Remove --wait long argument, keep only -w short form

ebe47b0

Co-authored-by: iandennismiller <90502+iandennismiller@users.noreply.github.com>

iandennismiller closed this Feb 18, 2026

iandennismiller mentioned this pull request Feb 18, 2026

New argument --kill-wait controls maximum time earlyoom waits for a killed process to exit #366

Open

Conversation

iandennismiller commented Feb 17, 2026

Uh oh!

rfjakob commented Feb 17, 2026

Uh oh!

iandennismiller commented Feb 17, 2026

Uh oh!

rfjakob commented Feb 17, 2026

Uh oh!

iandennismiller commented Feb 17, 2026

Uh oh!

rfjakob commented Feb 17, 2026

Uh oh!

iandennismiller commented Feb 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rfjakob left a comment

Choose a reason for hiding this comment

Uh oh!

rfjakob Feb 17, 2026

Choose a reason for hiding this comment

Uh oh!

rfjakob Feb 17, 2026

Choose a reason for hiding this comment

Uh oh!

rfjakob Feb 17, 2026

Choose a reason for hiding this comment

Uh oh!

rfjakob Feb 17, 2026

Choose a reason for hiding this comment

Uh oh!

rfjakob left a comment

Choose a reason for hiding this comment

Uh oh!

iandennismiller commented Feb 18, 2026

Uh oh!

iandennismiller commented Feb 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

iandennismiller commented Feb 17, 2026 •

edited

Loading