I was just witnessing a vip-manager2 autopkgtest run take down a whole
Jenkins node:
https://jengus.postgresql.org/job/vip-manager2-autopkgtest/65/architecture=amd64,distribution=sid/console
20:37:06 2026-10-13T18:37:03.839Z ␛[34mINFO␛[0m Removing address 10.0.2.123/32 on ceth1
20:37:06 Ncat: Version 7.991 ( https://nmap.org/ncat )
20:37:06 Ncat: TIMEOUT.
20:37:06 ␛[0;32m### You've reached the end of the script, all "tests" have successfully been passed! ###␛[0m
20:37:06
20:37:06 Session terminated, killing shell...autopkgtest [18:37:06]: ERROR: Received signal 15, cleaning up...
20:37:06 E: Caught signal ‘Terminated’: terminating immediately
20:37:06 FATAL: command execution failed
20:37:06 java.nio.channels.ClosedChannelException
The same failure is also visible on ci.debian.net, albeit less loudly:
https://ci.debian.net/packages/v/vip-manager2/testing/amd64/76553962/
155s Ncat: Connection from 10.0.2.123:35736.
155s ### You've reached the end of the script, all "tests" have successfully been passed! ###
155s
156s Session terminated, killing shell...autopkgtest [13:25:38]: test clientcert_test: -----------------------]
▾ test clientcert_test: test results
157s autopkgtest [13:25:39]: test clientcert_test: - - - - - - - - - - results - - - - - - - - - -
157s clientcert_test FAIL non-zero exit status 143
So, something in the test sends a kill signal to the process group and
autopkgtest just dies.
There are bugs on several layers:
* ncat being killed should not take the whole process group down (#1150845)
* autopkgtest should isolate itself against that happening (this bug)
* and on apt.postgresql.org, the "newpid" isolation layer should also
isolate (https://github.com/df7cb/newpid/commit/2e37c0be35563da7835f3dd57f0256a8ce9edf1c)
Afaict the proper fix is calling setsid(). (If interactive sessions
are involved, it's more complex, see the newpid fix.)
Christoph