Fix PR_SET_PDEATHSIG results in Broken pipe (#2395) #3353
Conversation
|
Hm, can't we just not set PR_SET_PDEATHSIG? |
|
We could and it would solve the ssh issue. However, I'm not sure whether the child cleanup will work in all cases then. As the PR_SET_PDEATHSIG was explicitly introduced to guarantee the proper cleanup, I was assuming it is still needed. I will check whether it has become redundant in the meantime. |
The ssh client is lazily started by the first worker thread, that requires a ssh connection. To avoid the ssh client to be killed, when the worker process is stopped, do not set PR_SET_PDEATHSIG.
tbsmoest
force-pushed
the
priv_tobias_pr_set_deathsig-1.4
branch
from
f174188
to
3e34722
Compare
|
It seems, that it is safe to not set PR_SET_PDEATHSIG for the SSH fork. The signals SIGINT and SIGTERM are forwarded. SIGKILL is not a problem either, as all children are killed, too. SIGSTOP causes issues with orphaned children when it is used in combination with SIGINT, but that is not solved by PR_SET_PDEATHSIG either. I have updated the PR accordingly. |
|
@edolstra, I have removed my initial solution and now simply deactivated the PR_SET_PDEATHSIG for ssh. I this ok with you? |
|
Thanks! |
|
@chpatrick made a backport of this fix for |
Fixes: #2395
The ssh client is lazily started by the first worker thread, that
requires a ssh connection. To avoid the ssh client to be killed, when
the worker process is stopped, start the ssh client in a long-lived
synchronous worker thread.
Add the class SyncWorker, that provides a single thread in which tasks
can be executed synchronously.