Repository navigation
feat(webserver): customize shutdown with new gracefulShutdown option - #34130
Conversation
Co-authored-by: Dmitry Gozman <dgozman@gmail.com> Signed-off-by: Simon Knott <info@simonknott.de>
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
| | `stdout` | If `"pipe"`, it will pipe the stdout of the command to the process stdout. If `"ignore"`, it will ignore the stdout of the command. Default to `"ignore"`. | | ||
| | `stderr` | Whether to pipe the stderr of the command to the process stderr or ignore it. Defaults to `"pipe"`. | | ||
| | `timeout` | How long to wait for the process to start up and be available in milliseconds. Defaults to 60000. | | ||
| | `kill` | How to shut down the process gracefully. If unspecified, the process group is forcefully `SIGKILL`ed. If set to `{ SIGINT: 500 }`, the process group is sent a `SIGINT` signal, followed by `SIGKILL` if it doesn't exit within 500ms. You can also use `SIGTERM` instead. A `0` timeout means no `SIGKILL` will be sent. Windows doesn't support `SIGINT` and `SIGTERM` signals, so this option is ignored. | |
There was a problem hiding this comment.
I'd split it into 2 fields{ signal: SIGINT', timeout 5000 } to make it clear that only one signal can be specified and what the value semantics is.
There was a problem hiding this comment.
Should we also rename it from kill to shutdown or gracefulShutdown? At the moment, kill: undefined means "KILL immediately", which could be surprising to some.
There was a problem hiding this comment.
Should we also rename it from
killtoshutdownorgracefulShutdown?
I like gracefulShutdown more. I'm sure there will be more opinions at the API review 😁
| killSet.delete(killProcessAndCleanup); | ||
| removeProcessHandlersIfNeeded(); | ||
| options.log(`[pid=${spawnedProcess.pid}] <kill>`); | ||
| if (spawnedProcess.pid && !spawnedProcess.killed && !processClosed) { |
There was a problem hiding this comment.
Previously we wouldn't send the signal again if it has already been received, I believe it shouldn't matter for SIGKILL handling, but would be nice to double check.
There was a problem hiding this comment.
This change was needed for when we used launchedProcess.kill(). We don't use that anymore, so i'll see if I can remove this change.
This comment has been minimized.
This comment has been minimized.
| attemptToGracefullyClose: () => Promise.reject(), | ||
| attemptToGracefullyClose: async () => { | ||
| if (process.platform === 'win32') | ||
| throw new Error('Graceful shutdown is not supported on Windows'); |
There was a problem hiding this comment.
Windows users will now always see ...<gracefully close start> followed by <will force kill> in the logs. Not a big deal, but this was not the case before as we never announced graceful shutdown there. Maybe keep using kill on Windows ?
There was a problem hiding this comment.
Don't think it makes any observable difference because log is a no-op anyways (see L128). So it's purely a code complexity question. If we assigned _killProcess depending on the platform, I worry that'd make us split execution more, and make this slightly harder to reason about I think. So i'll keep it as is for now.
Simon Knott (Skn0tt)
changed the title
kill optiongracefulShutdown option
Test results for "tests 1"4 flaky37513 passed, 653 skipped Merge workflow run. |
|
The difference between SIGINT and SIGTERM on Linux, which perhaps should be mentioned in the documentation, is that SIGINT is what is sent to the process running in the current terminal when you press Ctrl+C in that terminal. SIGTERM is the default signal sent by For some programs, such as programming languages with a REPL, SIGINT is captured and treated as "return to the REPL prompt" rather than "terminate process and return to the terminal prompt". So the documentation should probably list SIGTERM as the preferred method, and give SIGINT as an alternative to SIGTERM, since SIGTERM is the signal that more closely matches the intent of |
|
Thanks for the context! Would you be willing to contribute those details in a docs PR? That'd be wonderful. |
3 participants
We had this change up earlier, but it staled because I was unsure what PID to signal. After doing some more research on it, it appears that both for
SIGINTandSIGTERMit's appropriate to send to the entire process group. So this PR is a reopening of #33379 with those learnings.Closes #33377, #18209