Commit Graph

73 Commits (fe26b709025e92124d27aff3728919fe6517b60a)

Author SHA1 Message Date
Alex Willmer 4996ec2f09 ansible_mitogen: Fix "filedescriptor out of range in select()" in WorkerProcess
`mitogen.parent.POLLER_LIGHTWEIGHT` will normally be `PollPoller`, falling
back to `EpollPoller`, `KqueuePoller`, or `Poller`.

Fixes #957

Co-authored-by: Luca Berruti <nadirio@gmail.com>
Co-authored-by: Philippe Kueck <bqobccy6ejnq2bqvmebqiwqha4cs4@protected32.unixadm.org>
7 months ago
Alex Willmer 1871f2a9b1 Remove vendored mitogen.compat.simplejson
Python 2.6 added json to the stdlib. We no longer support Python <= 2.7 in
Mitogen 0.3.x, so this fallback is unneeded complexity. Fixes #659
2 years ago
Alex Willmer 31b3a4eb4a ansible_mitogen: Standardise __future__ imports to match Ansible
Some modules additionally enable unicode_literals (which Ansible doesn't do).
I've chosen not to change that, for now.
3 years ago
David Wilson e02be89879 issue #625: ignore SIGINT within MuxProcess
Without this, MuxProcess will start dying too early, before Ansible /
TaskQueueManager.cleanup() has a chance to wait on worker processes.
That would allow WorkerProcess to see ECONNREFUSED from the MuxProcess
socket much more easily.
5 years ago
David Wilson 67759371f9 issue #615: ensure 4GB max_message_size is configured for task workers.
This 4GB limit was already set for MuxProcess and inherited by all
descendents including the context running on the target host, but it was
not applied to the WorkerProcess router.

That explains why the error from the ticket is being raised by the
router within the WorkerProcess rather than the router on the original
target.
5 years ago
David Wilson 3b63da670f Fix up another handful of LGTM errors. 5 years ago
David Wilson 5298e87548 Split out and make readable more log messages across both packages 5 years ago
David Wilson 0f23a90d50 ansible: log affinity assignments 5 years ago
David Wilson 4f051a38a7 ansible: improve docstring 5 years ago
David Wilson 5811909c8d [linear2] simplify _listener_for_name() 5 years ago
David Wilson c68dbdd569 ansible: stop relying on SIGTERM to shut down service pool
It's no longer necessary, since connection attempts are no longer truly
blocking. When CTRL+C is hit in the top-level process, broker will begin
shutdown, which will cancel all pending connection attempts, causing
pool threads to wake. The pool can't block during shutdown anymore.
5 years ago
David Wilson f4ca926b21 ansible: cleanup various docstrings 5 years ago
David Wilson edde251d58 issue #549: ansible: reduce risk by capping RLIM_INFINITY 5 years ago
David Wilson d408caccf5 issue #573: guard against a forked top-level Ansible process
See comment.
5 years ago
David Wilson 3ceac2c9ed [linear2] simplify ClassicWorkerModel and fix repeat initialization
"self.initialized = False" slipped in a few days ago, on second thoughts
that flag is not needed at all, by simply rearranging ClassicWorkerModel
to have a regular constructor.

This hierarchy is still squishy, it needs more love. Remaining
MuxProcess class attributes should eliminated.
5 years ago
David Wilson 395b03a77d issue #549: fix setrlimit() crash and hard-wire OS X default
OS X advertised unlimited, but really it means kern.maxfilesperproc.
5 years ago
David Wilson 619f4dee07 [linear2] merge fallout: restore optimization from #491 / 7b129e857 5 years ago
David Wilson 75d179e4b9 remove unused imports flagged by lgtm 5 years ago
David Wilson eeb7150f24 issue #549: increase open file limit automatically if possible
While catching every possible case where "open file limit exceeded" is
not possible, we can at least increase the soft limit to the available
hard limit without any user effort.

Do this in Ansible top-level process, even though we probably only need
it in the MuxProcess. It seems there is no reason this could hurt
5 years ago
David Wilson acab26d796 ansible: improve process.py docs 5 years ago
David Wilson 4dfbe82e76 tests: hide ugly error during Ansible tests 5 years ago
David Wilson 108015aa22 ansible: gracefully handle failure to connect to MuxProcess
It's possible to hit an ugly exception during early CTRL+C
5 years ago
David Wilson bf1f3682aa ansible: pin per-CPU muxes to their corresponding CPU
This slightly breaks the old scheme, in that CPU 1 may now end up with a
mux and the top-level process pinned to it.
5 years ago
David Wilson dc9f4e89e6 ansible: reap mux processes on shut down
Previously we exitted without calling waitpid(), which meant the
top-level process struct rusage did not reflect the resource usage
consumed by the multiplexer processes.

Existing benchmarks are made using perf so this never created a problem,
but it could be confusing to others using the "time" command, and also
allows logging the final exit status of the process.
5 years ago
David Wilson 1fca0b7a94 [linear2] fix MuxProcess test fixture and some merge fallout 5 years ago
David Wilson 0f63ca4c68 Make setting affinity optional. 5 years ago
David Wilson 9035884c77 ansible: abstract worker process model.
Move all details of broker/router setup out of connection.py, instead
deferring it to a WorkerModel class exported by process.py via
get_worker_model(). The running strategy can override the configured
worker model via _get_worker_model().

ClassicWorkerModel is installed by default, which implements the
extension's existing process model.

Add optional support for the third party setproctitle module, so
children have pretty names in ps output.

Add optional support for per-CPU multiplexers to classic runs.
5 years ago
David Wilson 300f8b2ff9 ansible: fixturize creation of MuxProcess
This relies on the previous commit resetting global variables.

Update clean_shutdown() to handle duplicate calls, due to tests
repeatedly installing it.
5 years ago
David Wilson 26b6333787 [stream-refactor] fix unix.Listener construction 5 years ago
David Wilson 7dacb68eeb issue #552: include process identity in log messages. 6 years ago
David Wilson 1f77d24bec Update copyright year everywhere. 6 years ago
David Wilson 7ff4e6694c issue #536: rework how 2.3-compatible simplejson is served
Regardless of the version of simplejson loaded in the master, load up
the ModuleResponder cache with our 2.4-compatible version.

To cope with simplejson being loaded due to modules like ec2_group that
try to import it before importing 'json', also update target.py to
remove it from the whitelist if a local 'json' module import succeeds.
6 years ago
David Wilson 05b1ccb658 ansible: stash PID files in CWD if requested for debugging. 6 years ago
David Wilson eb67fbe9d2 ansible: double the default pool size.
Tempted to push this up to 64, but let's do it incrementally just in
case.
6 years ago
David Wilson 4531338b12 ansible: document and make affinity stuff portable to non-Linux
Portable as in does nothing for the time at least for now.
6 years ago
David Wilson c6d5aa29ba ansible: new multiplexer/workers configuration
Following on from 152effc26c9a5918cb7ead7a97fe7fa7f81b6764,

* Pin mux to CPU 0
* Pin top-level CPU 1
* Pin workers sequentially to CPU 2..n

Nets 19.5% improvement on issue_140__thread_pileup.yml when targetting
64 Docker containers on the same 8 core/16 thread machine.

Before (prior to last scheme, no affinity at all):

    2294528.731458      task-clock (msec)         #    6.443 CPUs utilized
        10,429,745      context-switches          #    0.005 M/sec
         2,049,618      cpu-migrations            #    0.893 K/sec
         8,258,952      page-faults               #    0.004 M/sec
 5,532,719,253,824      cycles                    #    2.411 GHz                      (83.35%)
 3,267,471,616,230      instructions              #    0.59  insn per cycle
                                                  #    1.22  stalled cycles per insn  (83.35%)
   662,006,455,943      branches                  #  288.515 M/sec                    (83.33%)
    39,453,895,977      branch-misses             #    5.96% of all branches          (83.37%)

     356.148064576 seconds time elapsed

After:

    2226463.958975      task-clock (msec)         #    7.784 CPUs utilized
         9,831,466      context-switches          #    0.004 M/sec
           180,065      cpu-migrations            #    0.081 K/sec
         5,082,278      page-faults               #    0.002 M/sec
 5,592,548,587,259      cycles                    #    2.512 GHz                      (83.35%)
 3,135,038,855,414      instructions              #    0.56  insn per cycle
                                                  #    1.32  stalled cycles per insn  (83.32%)
   636,397,509,232      branches                  #  285.833 M/sec                    (83.30%)
    39,135,441,790      branch-misses             #    6.15% of all branches          (83.35%)

     286.036681644 seconds time elapsed
6 years ago
David Wilson 1b909e8697 ansible: pin connection multiplexer to a single core
Nets a reliable 8% improvement in issue_140__thread_pileup.yml when
targetting 64 Docker containers on the same 8 core/16 thread machine.

Before:
    2294528.731458      task-clock (msec)         #    6.443 CPUs utilized
        10,429,745      context-switches          #    0.005 M/sec
         2,049,618      cpu-migrations            #    0.893 K/sec
         8,258,952      page-faults               #    0.004 M/sec
 5,532,719,253,824      cycles                    #    2.411 GHz                      (83.35%)
 4,001,276,805,120      stalled-cycles-frontend   #   72.32% frontend cycles idle     (83.30%)
 2,024,159,442,463      stalled-cycles-backend    #   36.59% backend cycles idle      (66.65%)
 3,267,471,616,230      instructions              #    0.59  insn per cycle
                                                  #    1.22  stalled cycles per insn  (83.35%)
   662,006,455,943      branches                  #  288.515 M/sec                    (83.33%)
    39,453,895,977      branch-misses             #    5.96% of all branches          (83.37%)

     356.148064576 seconds time elapsed

After:
    2208247.938562      task-clock (msec)         #    6.735 CPUs utilized
         8,489,840      context-switches          #    0.004 M/sec
         1,432,967      cpu-migrations            #    0.649 K/sec
         7,508,957      page-faults               #    0.003 M/sec
 5,477,293,750,357      cycles                    #    2.480 GHz                      (83.31%)
 3,984,360,350,811      stalled-cycles-frontend   #   72.74% frontend cycles idle     (83.32%)
 1,976,646,418,711      stalled-cycles-backend    #   36.09% backend cycles idle      (66.64%)
 3,196,197,480,792      instructions              #    0.58  insn per cycle
                                                  #    1.25  stalled cycles per insn  (83.36%)
   648,247,332,967      branches                  #  293.557 M/sec                    (83.35%)
    39,004,881,070      branch-misses             #    6.02% of all branches          (83.37%)

     327.876903668 seconds time elapsed
6 years ago
David Wilson 84944a9a61 ansible: ensure MuxProcess MITOGEN_PROFILING results reach disk.
This has been broken for quite some time.
6 years ago
David Wilson be6ab52fe1 issue #488: fix shutdown damage caused in 6ca2677de5
os._exit() subverted calm shutdown, meaning unix.Listener never had a
chance to cleanup its socket.

Move unix.Listener socket cleanup into its class so it is automatic
during shutdown, rather than cutpasted for each consumer.

Disable the watcher thread in the MuxProcess, it is useless.

Add .sock extension to /tmp/mitogen_unix_*, so we can write a test.
6 years ago
David Wilson dd30a907ce issue #477: promote setup_gil() to mitogen.utils
This is since ansible_mitogen/process.py is 2.6-only, and I want to use
setup_gil() in 2.4 code.
6 years ago
David Wilson a48ee3a536 issue #477: vendorize the last 2.4-compatible simplejson
This is in part so image_prep can run against an ancient CentOS 5 image
without any upfront help, and in part simply because it's very easy to
support.
6 years ago
David Wilson 59dd0dc814 issue #477: serve up junk ansible/__init__.py just like Ansible. 6 years ago
David Wilson 6ca2677de5 ansible: fix test failure during process exit.
======================================================================
ERROR: tests.connection_test (unittest2.loader._FailedTest)
----------------------------------------------------------------------
Traceback (most recent call last):
ImportError: Failed to import test module: tests.connection_test
Traceback (most recent call last):
  File "/home/dmw/src/mitogen/.venv/local/lib/python2.7/site-packages/unittest2/loader.py", line 456, in _find_test_path
    module = self._get_module_from_name(name)
  File "/home/dmw/src/mitogen/.venv/local/lib/python2.7/site-packages/unittest2/loader.py", line 395, in _get_module_from_name
    __import__(name)
RuntimeError: not holding the import lock
6 years ago
David Wilson 4bdf60326c issue #424: ansible: make put_file() raise AnsibleFileNotFound 6 years ago
David Wilson e647adc62e ansible: copy GIL change from linear2 branch.
Reduces runtime by 25% given 100 25ms SSH targets:

    ANSIBLE_STRATEGY=mitogen \
    MITOGEN_POOL_SIZE=100 \
    /usr/bin/time -l ansible k3-x100 -m shell -a hostname

Before:
           39.56 real        35.29 user        17.24 sys
      59600896  maximum resident set size
       1784252  page reclaims
          9016  messages sent
         10382  messages received
         18774  voluntary context switches
        770070  involuntary context switches

After:
           29.79 real        22.10 user        11.77 sys
      59281408  maximum resident set size
       1725268  page reclaims
          8582  messages sent
          9959  messages received
         14582  voluntary context switches
         75280  involuntary context switches
6 years ago
David Wilson 2647f73501 ansible: bump UNIX listener default backlog, and set it to match forks.
The connection multiplexer can expect to not be scheduled at least until
every $forks worker processes has attempted a connection, so the backlog
must be able to hold every worker.
6 years ago
David Wilson 8ab11f415f ansible: better support for diagnosing hangs
* Always enable the faulthandler module in the top-level process if it
  is available.
* Make MITOGEN_DUMP_THREAD_STACKS interval configurable, to better
  handle larger runs.
* Add docs subsection on diagnosing hangs.

Conflicts:
	ansible_mitogen/process.py
6 years ago
David Wilson e18396d54d ansible: enable profiling by default!
Thankfully this never made it into a release
6 years ago
David Wilson 9e572a7939 ansible: fix duplicate MuxProcess socket write.
The while: loop was necessary due to some cutpaste further on down the
file.
6 years ago
David Wilson 053c594d65 ansible: prevent logs spamming user console on exit.
Closes #331.
6 years ago