view hgext/remotefilelog/debugcommands.py @ 42354:dbd0fcca6dfc

discovery: slowly increase sampling size

Some pathological discovery runs can requires many roundtrip. When this happens
things can get very slow.

To make the algorithm more resilience again such pathological case. We slowly
increase the sample size with each roundtrip (+5%). This will have a negligible
impact on "normal" discovery with few roundtrips, but a large positive impact of
case with many roundtrips. Asking more question per roundtrip helps to reduce
the undecided set faster. Instead of reducing the undecided set a linear speed
(in the worst case), we reduce it as a guaranteed (small) exponential rate. The
data below show this slow ramp up in sample size:

round trip    |    1 |     5 |    10 |    20 |     50 |     100 |       130 |
sample size   |  200 |   254 |   321 |   517 |  2 199 |  25 123 |   108 549 |
covered nodes |  200 | 1 357 | 2 821 | 7 031 | 42 658 | 524 530 | 2 276 755 |

To be a bit more concrete, lets take a very pathological case as an example. We
are doing discovery from a copy of Mozilla-try to a more recent version of
mozilla-unified. Mozilla-unified heads are unknown to the mozilla-try repo and
there are over 1 million "missing" changesets. (the discovery is "local" to
avoid network interference)

Without this change, the discovery:
- last 1858 seconds (31 minutes),
- does 1700 round trip,
- asking about 340 000 nodes.

With this change, the discovery:
- last 218 seconds (3 minutes, 38 seconds a -88% improvement),
- does 94 round trip (-94%),
- asking about 344 211 nodes (+1%).

Of course, this is an extreme case (and 3 minutes is still slow). However this
give a good example of how this sample size increase act as a safety net
catching any bad situations.

We could image a steeper increase than 5%. For example 10% would give the
following number:

round trip    |    1 |     5 |    10 |     20 |     50  |        75 |        100 |
sample size   |  200 |   321 |   514 |  1 326 |  23 060 |   249 812 |  2 706 594 |
covered nodes |  200 | 1 541 | 3 690 | 12 671 | 251 871 | 2 746 254 | 29 770 966 |

In parallel, it is useful to understand these pathological cases and improve
them. However the current change provides a general purpose safety net to smooth
the impact of pathological cases.

To avoid issue with older http server, the increase in sample size only occurs
if the protocol has not limit on command argument size.

# debugcommands.py - debug logic for remotefilelog
#
# Copyright 2013 Facebook, Inc.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
from __future__ import absolute_import

import hashlib
import os
import zlib

from mercurial.node import bin, hex, nullid, short
from mercurial.i18n import _
from mercurial import (
    error,
    filelog,
    node as nodemod,
    pycompat,
    revlog,
)
from . import (
    constants,
    datapack,
    extutil,
    fileserverclient,
    historypack,
    repack,
    shallowutil,
)

def debugremotefilelog(ui, path, **opts):
    decompress = opts.get(r'decompress')

    size, firstnode, mapping = parsefileblob(path, decompress)

    ui.status(_("size: %d bytes\n") % (size))
    ui.status(_("path: %s \n") % (path))
    ui.status(_("key: %s \n") % (short(firstnode)))
    ui.status(_("\n"))
    ui.status(_("%12s => %12s %13s %13s %12s\n") %
              ("node", "p1", "p2", "linknode", "copyfrom"))

    queue = [firstnode]
    while queue:
        node = queue.pop(0)
        p1, p2, linknode, copyfrom = mapping[node]
        ui.status(_("%s => %s  %s  %s  %s\n") %
            (short(node), short(p1), short(p2), short(linknode), copyfrom))
        if p1 != nullid:
            queue.append(p1)
        if p2 != nullid:
            queue.append(p2)

def buildtemprevlog(repo, file):
    # get filename key
    filekey = nodemod.hex(hashlib.sha1(file).digest())
    filedir = os.path.join(repo.path, 'store/data', filekey)

    # sort all entries based on linkrev
    fctxs = []
    for filenode in os.listdir(filedir):
        if '_old' not in filenode:
            fctxs.append(repo.filectx(file, fileid=bin(filenode)))

    fctxs = sorted(fctxs, key=lambda x: x.linkrev())

    # add to revlog
    temppath = repo.sjoin('data/temprevlog.i')
    if os.path.exists(temppath):
        os.remove(temppath)
    r = filelog.filelog(repo.svfs, 'temprevlog')

    class faket(object):
        def add(self, a, b, c):
            pass
    t = faket()
    for fctx in fctxs:
        if fctx.node() not in repo:
            continue

        p = fctx.filelog().parents(fctx.filenode())
        meta = {}
        if fctx.renamed():
            meta['copy'] = fctx.renamed()[0]
            meta['copyrev'] = hex(fctx.renamed()[1])

        r.add(fctx.data(), meta, t, fctx.linkrev(), p[0], p[1])

    return r

def debugindex(orig, ui, repo, file_=None, **opts):
    """dump the contents of an index file"""
    if (opts.get(r'changelog') or
        opts.get(r'manifest') or
        opts.get(r'dir') or
        not shallowutil.isenabled(repo) or
        not repo.shallowmatch(file_)):
        return orig(ui, repo, file_, **opts)

    r = buildtemprevlog(repo, file_)

    # debugindex like normal
    format = opts.get('format', 0)
    if format not in (0, 1):
        raise error.Abort(_("unknown format %d") % format)

    generaldelta = r.version & revlog.FLAG_GENERALDELTA
    if generaldelta:
        basehdr = ' delta'
    else:
        basehdr = '  base'

    if format == 0:
        ui.write(("   rev    offset  length " + basehdr + " linkrev"
                  " nodeid       p1           p2\n"))
    elif format == 1:
        ui.write(("   rev flag   offset   length"
                  "     size " + basehdr + "   link     p1     p2"
                  "       nodeid\n"))

    for i in r:
        node = r.node(i)
        if generaldelta:
            base = r.deltaparent(i)
        else:
            base = r.chainbase(i)
        if format == 0:
            try:
                pp = r.parents(node)
            except Exception:
                pp = [nullid, nullid]
            ui.write("% 6d % 9d % 7d % 6d % 7d %s %s %s\n" % (
                    i, r.start(i), r.length(i), base, r.linkrev(i),
                    short(node), short(pp[0]), short(pp[1])))
        elif format == 1:
            pr = r.parentrevs(i)
            ui.write("% 6d %04x % 8d % 8d % 8d % 6d % 6d % 6d % 6d %s\n" % (
                    i, r.flags(i), r.start(i), r.length(i), r.rawsize(i),
                    base, r.linkrev(i), pr[0], pr[1], short(node)))

def debugindexdot(orig, ui, repo, file_):
    """dump an index DAG as a graphviz dot file"""
    if not shallowutil.isenabled(repo):
        return orig(ui, repo, file_)

    r = buildtemprevlog(repo, os.path.basename(file_)[:-2])

    ui.write(("digraph G {\n"))
    for i in r:
        node = r.node(i)
        pp = r.parents(node)
        ui.write("\t%d -> %d\n" % (r.rev(pp[0]), i))
        if pp[1] != nullid:
            ui.write("\t%d -> %d\n" % (r.rev(pp[1]), i))
    ui.write("}\n")

def verifyremotefilelog(ui, path, **opts):
    decompress = opts.get(r'decompress')

    for root, dirs, files in os.walk(path):
        for file in files:
            if file == "repos":
                continue
            filepath = os.path.join(root, file)
            size, firstnode, mapping = parsefileblob(filepath, decompress)
            for p1, p2, linknode, copyfrom in mapping.itervalues():
                if linknode == nullid:
                    actualpath = os.path.relpath(root, path)
                    key = fileserverclient.getcachekey("reponame", actualpath,
                                                       file)
                    ui.status("%s %s\n" % (key, os.path.relpath(filepath,
                                                                path)))

def _decompressblob(raw):
    return zlib.decompress(raw)

def parsefileblob(path, decompress):
    f = open(path, "rb")
    try:
        raw = f.read()
    finally:
        f.close()

    if decompress:
        raw = _decompressblob(raw)

    offset, size, flags = shallowutil.parsesizeflags(raw)
    start = offset + size

    firstnode = None

    mapping = {}
    while start < len(raw):
        divider = raw.index('\0', start + 80)

        currentnode = raw[start:(start + 20)]
        if not firstnode:
            firstnode = currentnode

        p1 = raw[(start + 20):(start + 40)]
        p2 = raw[(start + 40):(start + 60)]
        linknode = raw[(start + 60):(start + 80)]
        copyfrom = raw[(start + 80):divider]

        mapping[currentnode] = (p1, p2, linknode, copyfrom)
        start = divider + 1

    return size, firstnode, mapping

def debugdatapack(ui, *paths, **opts):
    for path in paths:
        if '.data' in path:
            path = path[:path.index('.data')]
        ui.write("%s:\n" % path)
        dpack = datapack.datapack(path)
        node = opts.get(r'node')
        if node:
            deltachain = dpack.getdeltachain('', bin(node))
            dumpdeltachain(ui, deltachain, **opts)
            return

        if opts.get(r'long'):
            hashformatter = hex
            hashlen = 42
        else:
            hashformatter = short
            hashlen = 14

        lastfilename = None
        totaldeltasize = 0
        totalblobsize = 0
        def printtotals():
            if lastfilename is not None:
                ui.write("\n")
            if not totaldeltasize or not totalblobsize:
                return
            difference = totalblobsize - totaldeltasize
            deltastr = "%0.1f%% %s" % (
                (100.0 * abs(difference) / totalblobsize),
                ("smaller" if difference > 0 else "bigger"))

            ui.write(("Total:%s%s  %s (%s)\n") % (
                "".ljust(2 * hashlen - len("Total:")),
                ('%d' % totaldeltasize).ljust(12),
                ('%d' % totalblobsize).ljust(9),
                deltastr
            ))

        bases = {}
        nodes = set()
        failures = 0
        for filename, node, deltabase, deltalen in dpack.iterentries():
            bases[node] = deltabase
            if node in nodes:
                ui.write(("Bad entry: %s appears twice\n" % short(node)))
                failures += 1
            nodes.add(node)
            if filename != lastfilename:
                printtotals()
                name = '(empty name)' if filename == '' else filename
                ui.write("%s:\n" % name)
                ui.write("%s%s%s%s\n" % (
                    "Node".ljust(hashlen),
                    "Delta Base".ljust(hashlen),
                    "Delta Length".ljust(14),
                    "Blob Size".ljust(9)))
                lastfilename = filename
                totalblobsize = 0
                totaldeltasize = 0

            # Metadata could be missing, in which case it will be an empty dict.
            meta = dpack.getmeta(filename, node)
            if constants.METAKEYSIZE in meta:
                blobsize = meta[constants.METAKEYSIZE]
                totaldeltasize += deltalen
                totalblobsize += blobsize
            else:
                blobsize = "(missing)"
            ui.write("%s  %s  %s%s\n" % (
                hashformatter(node),
                hashformatter(deltabase),
                ('%d' % deltalen).ljust(14),
                pycompat.bytestr(blobsize)))

        if filename is not None:
            printtotals()

        failures += _sanitycheck(ui, set(nodes), bases)
        if failures > 1:
            ui.warn(("%d failures\n" % failures))
            return 1

def _sanitycheck(ui, nodes, bases):
    """
    Does some basic sanity checking on a packfiles with ``nodes`` ``bases`` (a
    mapping of node->base):

    - Each deltabase must itself be a node elsewhere in the pack
    - There must be no cycles
    """
    failures = 0
    for node in nodes:
        seen = set()
        current = node
        deltabase = bases[current]

        while deltabase != nullid:
            if deltabase not in nodes:
                ui.warn(("Bad entry: %s has an unknown deltabase (%s)\n" %
                        (short(node), short(deltabase))))
                failures += 1
                break

            if deltabase in seen:
                ui.warn(("Bad entry: %s has a cycle (at %s)\n" %
                        (short(node), short(deltabase))))
                failures += 1
                break

            current = deltabase
            seen.add(current)
            deltabase = bases[current]
        # Since ``node`` begins a valid chain, reset/memoize its base to nullid
        # so we don't traverse it again.
        bases[node] = nullid
    return failures

def dumpdeltachain(ui, deltachain, **opts):
    hashformatter = hex
    hashlen = 40

    lastfilename = None
    for filename, node, filename, deltabasenode, delta in deltachain:
        if filename != lastfilename:
            ui.write("\n%s\n" % filename)
            lastfilename = filename
        ui.write("%s  %s  %s  %s\n" % (
            "Node".ljust(hashlen),
            "Delta Base".ljust(hashlen),
            "Delta SHA1".ljust(hashlen),
            "Delta Length".ljust(6),
        ))

        ui.write("%s  %s  %s  %d\n" % (
            hashformatter(node),
            hashformatter(deltabasenode),
            nodemod.hex(hashlib.sha1(delta).digest()),
            len(delta)))

def debughistorypack(ui, path):
    if '.hist' in path:
        path = path[:path.index('.hist')]
    hpack = historypack.historypack(path)

    lastfilename = None
    for entry in hpack.iterentries():
        filename, node, p1node, p2node, linknode, copyfrom = entry
        if filename != lastfilename:
            ui.write("\n%s\n" % filename)
            ui.write("%s%s%s%s%s\n" % (
                "Node".ljust(14),
                "P1 Node".ljust(14),
                "P2 Node".ljust(14),
                "Link Node".ljust(14),
                "Copy From"))
            lastfilename = filename
        ui.write("%s  %s  %s  %s  %s\n" % (short(node), short(p1node),
            short(p2node), short(linknode), copyfrom))

def debugwaitonrepack(repo):
    with extutil.flock(repack.repacklockvfs(repo).join('repacklock'), ''):
        return

def debugwaitonprefetch(repo):
    with repo._lock(repo.svfs, "prefetchlock", True, None,
                         None, _('prefetching in %s') % repo.origroot):
        pass