[draft] erts: kill spawned child processes on VM exit #9453
Conversation
|
|
CT Test Results 3 files 141 suites 49m 55s ⏱️ Results for commit dba896f. ♻️ This comment has been updated with latest results. To speed up review, make sure that you have read Contributing to Erlang/OTP and that all checks pass. See the TESTING and DEVELOPMENT HowTo guides for details about how to run test locally. Artifacts// Erlang/OTP Github Action Bot |
| } | ||
|
|
||
| static Eterm get_port_id(pid_t os_pid) | ||
| { | ||
| ErtsSysExitStatus est, *es; | ||
| Eterm port_id; | ||
| est.os_pid = os_pid; | ||
| es = hash_remove(forker_hash, &est); | ||
| es = hash_get(forker_hash, &est); | ||
| if (!es) return THE_NON_VALUE; | ||
| port_id = es->port_id; |
There was a problem hiding this comment.
This seems to preserve the original behavior of only sending exit_status back to callers which have requested it, since port_id is set conditionally by the caller and is still used in the guard around sending—not simply whether the os_pid exists in the hash table.
adamwight
force-pushed
the
aw-orphans
branch
5 times, most recently
from
9f87bc1 to
dba896f
Compare
|
Hello! I think that we can move forward with this. There is no need to have an option to disable it for now (unless our existing tests shows that it is needed...), but there needs to be testcases to test that it works as expected on both Unix and Windows. I do wonder however if we should send some other signal than |
Good point, TIL that sigkill is untrappable. Looking at I experimented a bit locally to see if |
Keep the mapping of all living child processes so that we'll be able to iterate over them to clean up, rather than only storing the children which have an associated port.
If the uds_fd connection to the parent BEAM is broken or closed, react
by killing all children and any descendants in the same process group.
A concise demonstration of the problem being solved is to run this
command with and without the patch, then kill the BEAM. Without the
patch, the "sleep" process will continue:
erl -noshell -eval 'os:cmd("sleep 60")'
To intentionally start a child process which can outlive BEAM
termination, give it a new process group for example by using
`setsid`:
erl -noshell -eval 'os:cmd("setsid sleep 60")'
FIXME: Not working yet—the grandchild must never get TERM? FIXME: Don't sleep for fixed intervals, use messages for timing.
adamwight
force-pushed
the
aw-orphans
branch
from
215fc42 to
ce88ab3
Compare
This is a very rough proof-of-concept for discussion, which ensures all children spawned with open_port are terminated along with the BEAM.
Will be discussed in https://erlangforums.com/t/open-port-and-zombie-processes