EINTR from communication with subprocess in allmydata/util/iputil.py _query

[davidsarah]

Reported by 'sickness' on irc:

#   Run
#     test_loadable ...                                                      [OK]
#     test_reloadable ... Node._startService failed, aborting
# [Failure instance: Traceback: <type 'exceptions.OSError'>: [Errno 4] Interrupted system call
# /usr/lib/python2.6/threading.py:497:__bootstrap
# /usr/lib/python2.6/threading.py:525:__bootstrap_inner
# /usr/lib/python2.6/threading.py:477:run
# --- <exception caught here> ---
# /usr/lib/python2.6/vendor-packages/twisted/python/threadpool.py:210:_worker
# /usr/lib/python2.6/vendor-packages/twisted/python/context.py:59:callWithContext
# /usr/lib/python2.6/vendor-packages/twisted/python/context.py:37:callWithContext
# /home/righdieg/allmydata-tahoe-1.8.2/src/allmydata/util/iputil.py:222:_synchronously_find_addresses_via_config
# /home/righdieg/allmydata-tahoe-1.8.2/src/allmydata/util/iputil.py:237:_query
# /usr/lib/python2.6/subprocess.py:689:communicate
# /usr/lib/python2.6/subprocess.py:1233:_communicate
# /usr/lib/python2.6/subprocess.py:1157:wait
# ]
# calling os.abort()

Possibly related: ​http://bugs.python.org/issue1068268 . It may be that the patch for that bug wasn't complete enough. EINTR failures are usually not very reproducible, but the fix is just to repeat the query until it works (or fails with a different error).

[sickness]

The OS is opensolaris snv134 64bit

$ uname -a

SunOS MYWORKPC 5.11 snv_134 i86pc i386 i86pc Solaris

$ psrinfo -pv

The physical processor has 2 virtual processors (0 1)

x86 (GenuineIntel? 1067A family 6 model 23 step 10 clock 2800 MHz)

Pentium(r) Dual-Core CPU E6300 @ 2.80GHz

$ isainfo -x

amd64: ssse3 cx16 mon sse3 sse2 sse fxsr mmx cmov amd_sysc cx8 tsc fpu

i386: ssse3 ahf cx16 mon sse3 sse2 sse fxsr mmx cmov sep cx8 tsc fpu

This is instead the tahoe version:

$ allmydata-tahoe-1.8.2/bin/tahoe --version

allmydata-tahoe: 1.8.2,

foolscap: 0.6.1,

pycryptopp: 0.5.29,

zfec: 1.4.22,

Twisted: 8.2.0,

Nevow: 0.10.0,

zope.interface: unknown,

python: 2.6.4,

platform: SunOS-5.11-i86pc-i386-32bit-ELF,

pyOpenSSL: 0.11,

simplejson: 2.0.9,

pycrypto: 2.3,

pyasn1: unknown,

mock: 0.7.0,

sqlite3: 2.4.1 [sqlite 3.6.17],

setuptools: 0.6c16dev3

[davidsarah]

Replying to sickness:

python: 2.6.4,

Hmm, that should have had the backported fix for ​http://bugs.python.org/issue1068268 . Oh well, we would need to work around it for earlier Pythons anyway.

[zooko]

Should we work-around this by catching OSError with errno==4 and retrying the subprocess?

[davidsarah]

Replying to zooko:

Should we work-around this by catching OSError with errno==4 and retrying the subprocess?

Yes, I believe so. We probably shouldn't retry forever, so let's retry 10 times. The try/except should cover lines 236 and 237 of iputil.py.

BTW, rather than 4 we should use errno.EINTR (I think this is defined on all platforms, even though EINTR is only really relevant on Unix).

Should _query return [] (i.e. no addresses) if the subprocess fails? Oh, I see that issue is #854 ('what to do when you can't find any IP address for yourself').

[zooko]

See #1988