setdiscovery: make progress on most connected groups each roundtrip
Consider history like this:
o
| o
| |
| o
| |
| o
|/
o
| o
| |
| o
| |
| o
|/
o
| o
| |
| o
| |
| o
|/
o
~
Assume the left mainline is available in the remote repo and the other
commits are only in the local repo. Also imagine that instead of 3
local branches with 3 commits on each, there are 1000 branches (the
number of commits on each doesn't matter much here). In such a
scenario, the current setdiscovery code will pick a sample size of 200
among these branches and ask the remote which of them it has. However,
the discovery for each such branch is completely independent of the
discovery for the others -- knowing whether the remote has a commit in
one branch doesn't give us any information about the other
branches. The discovery will therefore take at least 5 roundtrips
(maybe more depending on which commit in each linear chain was
sampled). Since the discovery for each branch is independent, there is
no reason to let one branch wait for another, so this patch makes it
so we sample at least as many commits as there are branches. It may
still happen (it's very likely, even) that we get multiple samples
from one branch and none from another, but that will even out over a
few rounds and I think this is still a big improvement.
Because of http header size limits, we still use the old behavior
unless experimental.httppostargs=true.
I've timed this by running `hg debugdiscovery mozilla-unified --debug` in the
mozilla-try repo. Both repos were local. Before this patch, last part
of the output was:
2249 total queries in 5276.4859s
elapsed time: 5276.652634 seconds
heads summary:
total common heads: 13
also local heads: 4
also remote heads: 8
both: 4
local heads: 28317
common: 4
missing: 28313
remote heads: 12
common: 8
unknown: 4
local changesets: 2014901
common: 530373
missing: 1484528
common heads:
1dad417c28ad 4a108e94d3e2 4d7ef530fffb 5350524bb654 777e60ca8853 7d97fafba271 9cd2ab4d0029 a55ce37217da d38398e5144e dcc6d7a0dc00 e09297892ada e24ec6070d7b fd559328eaf3
After this patch, the output was (including all the samples, since
there were so few now):
taking initial sample
query 2; still undecided: 1599476, sample size is: 108195
sampling from both directions
query 3; still undecided: 810922, sample size is: 194158
sampling from both directions
query 4; still undecided: 325882, sample size is: 137302
sampling from both directions
query 5; still undecided: 111459, sample size is: 74586
sampling from both directions
query 6; still undecided: 26805, sample size is: 23960
sampling from both directions
query 7; still undecided: 2549, sample size is: 2528
sampling from both directions
query 8; still undecided: 21, sample size is: 21
8 total queries in 24.5064s
elapsed time: 24.670051 seconds
heads summary:
total common heads: 13
also local heads: 4
also remote heads: 8
both: 4
local heads: 28317
common: 4
missing: 28313
remote heads: 12
common: 8
unknown: 4
local changesets: 2014901
common: 530373
missing: 1484528
common heads:
1dad417c28ad 4a108e94d3e2 4d7ef530fffb 5350524bb654 777e60ca8853 7d97fafba271 9cd2ab4d0029 a55ce37217da d38398e5144e dcc6d7a0dc00 e09297892ada e24ec6070d7b fd559328eaf3
Differential Revision: https://phab.mercurial-scm.org/D2647
#!/usr/bin/env python3
#
# Copyright 2018 Google LLC.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
"""Tool read primitive events from a pipe to produce a catapult trace.
Usage:
Terminal 1: $ catapipe.py /tmp/mypipe /tmp/trace.json
Terminal 2: $ HGCATAPULTSERVERPIPE=/tmp/mypipe hg root
<ctrl-c catapipe.py in Terminal 1>
$ catapult/tracing/bin/trace2html /tmp/trace.json # produce /tmp/trace.html
<open trace.html in your browser of choice; the WASD keys are very useful>
(catapult is located at https://github.com/catapult-project/catapult)
For now the event stream supports
START $SESSIONID ...
and
END $SESSIONID ...
events. Everything after the SESSIONID (which must not contain spaces)
is used as a label for the event. Events are timestamped as of when
they arrive in this process and are then used to produce catapult
traces that can be loaded in Chrome's about:tracing utility. It's
important that the event stream *into* this process stay simple,
because we have to emit it from the shell scripts produced by
run-tests.py.
Typically you'll want to place the path to the named pipe in the
HGCATAPULTSERVERPIPE environment variable, which both run-tests and hg
understand. To trace *only* run-tests, use HGTESTCATAPULTSERVERPIPE instead.
"""
from __future__ import absolute_import, print_function
import argparse
import json
import os
import timeit
_TYPEMAP = {
'START': 'B',
'END': 'E',
}
_threadmap = {}
# Timeit already contains the whole logic about which timer to use based on
# Python version and OS
timer = timeit.default_timer
def main():
parser = argparse.ArgumentParser()
parser.add_argument('pipe', type=str, nargs=1,
help='Path of named pipe to create and listen on.')
parser.add_argument('output', default='trace.json', type=str, nargs='?',
help='Path of json file to create where the traces '
'will be stored.')
parser.add_argument('--debug', default=False, action='store_true',
help='Print useful debug messages')
args = parser.parse_args()
fn = args.pipe[0]
os.mkfifo(fn)
try:
with open(fn) as f, open(args.output, 'w') as out:
out.write('[\n')
start = timer()
while True:
ev = f.readline().strip()
if not ev:
continue
now = timer()
if args.debug:
print(ev)
verb, session, label = ev.split(' ', 2)
if session not in _threadmap:
_threadmap[session] = len(_threadmap)
pid = _threadmap[session]
ts_micros = (now - start) * 1000000
out.write(json.dumps(
{
"name": label,
"cat": "misc",
"ph": _TYPEMAP[verb],
"ts": ts_micros,
"pid": pid,
"tid": 1,
"args": {}
}))
out.write(',\n')
finally:
os.unlink(fn)
if __name__ == '__main__':
main()