bug#22130: dmd: respawn test failures

bug#22130: dmd: respawn test failures

[Mark H Weaver]

In DMD, the "respawn.sh" test fails quite consistently on armhf, and has
sometimes failed on mips64el and x86_64 as well.  Here's a copy of
test-suite.log from a recent failure on armhf:

--8<---------------cut here---------------start------------->8---
===================================
   GNU dmd 0.2: ./test-suite.log
===================================

# TOTAL: 3
# PASS:  2
# SKIP:  0
# XFAIL: 0
# FAIL:  1
# XPASS: 0
# ERROR: 0

.. contents:: :depth: 2

FAIL: tests/respawn
===================

+ dmd --version
dmd (GNU dmd) 0.2
Copyright (C) 2013 Ludovic Court?s
Copyright (C) 2002, 2003 Wolfgang J?hrling
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
+ deco --version
deco (GNU dmd) 0.2
Copyright (C) 2013 Ludovic Court?s
Copyright (C) 2002, 2003 Wolfgang J?hrling
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
+ socket=t-socket-759
+ conf=t-conf-759
+ log=t-log-759
+ stamp=t-stamp-759
+ service1_pid=t-service1-pid-759
+ service2_pid=t-service2-pid-759
+ pid=t-pid-759
+ deco='deco -s t-socket-759'
+ trap 'cat t-log-759 || true ;
  rm -f t-socket-759 t-conf-759 t-stamp-759 t-log-759 t-pid-759 t-service1-pid-759 t-service2-pid-759 ;
  test -f t-pid-759 && kill `cat t-pid-759` || true ;
  test -f t-service1-pid-759 && kill `cat t-service1-pid-759` || true ;
  test -f t-service2-pid-759 && kill `cat t-service2-pid-759` || true ;
  rm -f t-service1-pid-759 t-service2-pid-759' EXIT
+ cat
+ rm -f t-pid-759
+ wait_for_file t-pid-759
+ i=0
+ test -f t-pid-759
+ test 0 -lt 20
+ sleep 0.3
+ dmd -I -s t-socket-759 -c t-conf-759 -l t-log-759 --pid=t-pid-759
++ expr 0 + 1
+ i=1
+ test -f t-pid-759
+ test 1 -lt 20
+ sleep 0.3
++ expr 1 + 1
+ i=2
+ test -f t-pid-759
+ test -f t-pid-759
++ cat t-pid-759
+ dmd_pid=774
+ kill -0 774
+ test -S t-socket-759
+ deco -s t-socket-759 status dmd
Started: (test1 test2 dmd)
Stopped: ()
+ grep started
+ deco -s t-socket-759 status test1
  It is started.
+ grep started
+ deco -s t-socket-759 status test2
  It is started.
+ test -f t-service1-pid-759
+ cat t-log-759
2015-12-09 16:15:18 Service dmd has been started.
2015-12-09 16:15:18 Service test1 has been started.
2015-12-09 16:15:18 Service test2 has been started.
2015-12-09 16:15:19 Started: (test1 test2 dmd)
2015-12-09 16:15:19 Stopped: ()
2015-12-09 16:15:19 Status of test1:
2015-12-09 16:15:19   It is started.
2015-12-09 16:15:19   Running value is 786.
2015-12-09 16:15:19   It is enabled.
2015-12-09 16:15:19   Provides (test1).
2015-12-09 16:15:19   Requires ().
2015-12-09 16:15:19   Conflicts with ().
2015-12-09 16:15:19   Will be respawned.
2015-12-09 16:15:20 Status of test2:
2015-12-09 16:15:20   It is started.
2015-12-09 16:15:20   Running value is 787.
2015-12-09 16:15:20   It is enabled.
2015-12-09 16:15:20   Provides (test2).
2015-12-09 16:15:20   Requires ().
2015-12-09 16:15:20   Conflicts with ().
2015-12-09 16:15:20   Will be respawned.
+ rm -f t-socket-759 t-conf-759 t-stamp-759 t-log-759 t-pid-759 t-service1-pid-759 t-service2-pid-759
+ test -f t-pid-759
+ true
+ test -f t-service1-pid-759
+ true
+ test -f t-service2-pid-759
+ true
+ rm -f t-service1-pid-759 t-service2-pid-759

--8<---------------cut here---------------end--------------->8---

      Mark

bug#22130: dmd: respawn test failures

[Ludovic Courtès]

Mark H Weaver <mhw@netris.org> skribis:

> + deco -s t-socket-759 status dmd
> Started: (test1 test2 dmd)
> Stopped: ()
> + grep started
> + deco -s t-socket-759 status test1
>   It is started.
> + grep started
> + deco -s t-socket-759 status test2
>   It is started.
> + test -f t-service1-pid-759
> + cat t-log-759

The problem here is that the PID file for ‘service1’ is not ready when
we ‘test -f’ it right after ‘deco status test1’ has told us the service
is started.  That’s because there’s a small window during which the
service is started but has not created its PID file yet.

Fixed in dmd commit f33f81f.

We’ll backport it in Guix or make a dmd release.

Thanks,
Ludo’.