view mercurial/verify.py @ 42745:4d20b1fe8a72

rust-discovery: using from Python code

As previously done in other topics, the Rust version is used if it's been
built.

The version fully in Rust of the partialdiscovery class has the performance
advantage over the Python version (actually using the Rust MissingAncestor) if
the undecided set is big enough. Otherwise no sampling occurs, and the
discovery is reasonably fast anyway.

Note: it's hard to predict the size of the initial undecided set, it can
depend on the kind of topological changes between the local and remote graphs.
The point of the Rust version is to make the bad cases acceptable.

More specifically, the performance advantages are:

- faster sampling, especially takefullsample()
- much faster addmissings() in almost all cases (see commit message in
  grandparent of the present changeset)
- no conversion cost of the undecided set at the interface between Rust and
  Python

== Measurements with big undecided sets

For an extreme example, discovery between mozilla-try and mozilla-unified
(over one million undecided revisions, same case as in dbd0fcca6dfc), we
get roughly a x2.5/x3 better performance:

Growing sample size (5% starting with 200): time goes down from
210 to 72 seconds.
Constant sample size of 200: time down from 1853 to 659 seconds.

With a sample size computed from number of roots and heads of the
undecided set (`respectsize` is `False`), here are perfdiscovery results:

Before ! wall 9.358729 comb 9.360000 user 9.310000 sys 0.050000 (median of 50)
After  ! wall 3.793819 comb 3.790000 user 3.750000 sys 0.040000 (median of 50)

In that later case, the sample sizes are routinely in the hundreds of
thousands of revisions. While still faster, the Rust iteration in
addmissings has less of an advantage than with smaller sample sizes, but
one sees addcommons becoming faster, probably a consequence of not having
to copy big sets back and forth.

This example is not a goal in itself, but it showcases several different
areas in which the process can become slow, due to different factors, and
how this full Rust version can help.

== Measurements with small undecided sets

In cases the undecided set is small enough than no sampling occurs,
the Rust version has a disadvantage at init if `targetheads` is really big
(some time is lost in the translation to Rust data structures),
and that is compensated by the faster `addmissings()`.

On a private repository with over one million commits, we still get a minor
improvement, of 6.8%:

  Before ! wall 0.593585 comb 0.590000 user 0.550000 sys 0.040000 (median of 50)
  After  ! wall 0.553035 comb 0.550000 user 0.520000 sys 0.030000 (median of 50)

What's interesting in that case is the first addinfo() at 180ms for Rust and
233ms for Python+C, mostly due to add_missings and the children cache
computation being done in less than 0.2ms on the Rust side vs over 40ms on the
Python side.

The worst case we have on hand is with mozilla-try, prepared with
discovery-helper.sh for 10 heads and depth 10, time goes up 2.2% on the median.
In this case `targetheads` is really huge with 165842 server heads.

  Before ! wall 0.823884 comb 0.810000 user 0.790000 sys 0.020000 (median of 50)
  After  ! wall 0.842607 comb 0.840000 user 0.800000 sys 0.040000 (median of 50)

If that would be considered a problem, more adjustments can be made, which are
prematurate at this stage: cooking special variants of methods of the inner
MissingAncestors object, retrieving local heads directly from Rust to avoid
the cost of conversion. Effort would probably be better spent at this point
improving the surroundings if needed.

Here's another data point with a smaller repository, pypy, where performance
is almost identical

 Before ! wall 0.015121 comb 0.030000 user 0.020000 sys 0.010000 (median of 186)
 After  ! wall 0.015009 comb 0.010000 user 0.010000 sys 0.000000 (median of 184)

Differential Revision: https://phab.mercurial-scm.org/D6430

# verify.py - repository integrity checking for Mercurial
#
# Copyright 2006, 2007 Matt Mackall <[e-mail]>
#
# 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 os

from .i18n import _
from .node import (
    nullid,
    short,
)

from . import (
    error,
    pycompat,
    revlog,
    util,
)

VERIFY_DEFAULT = 0
VERIFY_FULL = 1

def verify(repo, level=None):
    with repo.lock():
        v = verifier(repo, level)
        return v.verify()

def _normpath(f):
    # under hg < 2.4, convert didn't sanitize paths properly, so a
    # converted repo may contain repeated slashes
    while '//' in f:
        f = f.replace('//', '/')
    return f

class verifier(object):
    def __init__(self, repo, level=None):
        self.repo = repo.unfiltered()
        self.ui = repo.ui
        self.match = repo.narrowmatch()
        if level is None:
            level = VERIFY_DEFAULT
        self._level = level
        self.badrevs = set()
        self.errors = 0
        self.warnings = 0
        self.havecl = len(repo.changelog) > 0
        self.havemf = len(repo.manifestlog.getstorage(b'')) > 0
        self.revlogv1 = repo.changelog.version != revlog.REVLOGV0
        self.lrugetctx = util.lrucachefunc(repo.__getitem__)
        self.refersmf = False
        self.fncachewarned = False
        # developer config: verify.skipflags
        self.skipflags = repo.ui.configint('verify', 'skipflags')
        self.warnorphanstorefiles = True

    def _warn(self, msg):
        """record a "warning" level issue"""
        self.ui.warn(msg + "\n")
        self.warnings += 1

    def _err(self, linkrev, msg, filename=None):
        """record a "error" level issue"""
        if linkrev is not None:
            self.badrevs.add(linkrev)
            linkrev = "%d" % linkrev
        else:
            linkrev = '?'
        msg = "%s: %s" % (linkrev, msg)
        if filename:
            msg = "%s@%s" % (filename, msg)
        self.ui.warn(" " + msg + "\n")
        self.errors += 1

    def _exc(self, linkrev, msg, inst, filename=None):
        """record exception raised during the verify process"""
        fmsg = pycompat.bytestr(inst)
        if not fmsg:
            fmsg = pycompat.byterepr(inst)
        self._err(linkrev, "%s: %s" % (msg, fmsg), filename)

    def _checkrevlog(self, obj, name, linkrev):
        """verify high level property of a revlog

        - revlog is present,
        - revlog is non-empty,
        - sizes (index and data) are correct,
        - revlog's format version is correct.
        """
        if not len(obj) and (self.havecl or self.havemf):
            self._err(linkrev, _("empty or missing %s") % name)
            return

        d = obj.checksize()
        if d[0]:
            self._err(None, _("data length off by %d bytes") % d[0], name)
        if d[1]:
            self._err(None, _("index contains %d extra bytes") % d[1], name)

        if obj.version != revlog.REVLOGV0:
            if not self.revlogv1:
                self._warn(_("warning: `%s' uses revlog format 1") % name)
        elif self.revlogv1:
            self._warn(_("warning: `%s' uses revlog format 0") % name)

    def _checkentry(self, obj, i, node, seen, linkrevs, f):
        """verify a single revlog entry

        arguments are:
        - obj:      the source revlog
        - i:        the revision number
        - node:        the revision node id
        - seen:     nodes previously seen for this revlog
        - linkrevs: [changelog-revisions] introducing "node"
        - f:        string label ("changelog", "manifest", or filename)

        Performs the following checks:
        - linkrev points to an existing changelog revision,
        - linkrev points to a changelog revision that introduces this revision,
        - linkrev points to the lowest of these changesets,
        - both parents exist in the revlog,
        - the revision is not duplicated.

        Return the linkrev of the revision (or None for changelog's revisions).
        """
        lr = obj.linkrev(obj.rev(node))
        if lr < 0 or (self.havecl and lr not in linkrevs):
            if lr < 0 or lr >= len(self.repo.changelog):
                msg = _("rev %d points to nonexistent changeset %d")
            else:
                msg = _("rev %d points to unexpected changeset %d")
            self._err(None, msg % (i, lr), f)
            if linkrevs:
                if f and len(linkrevs) > 1:
                    try:
                        # attempt to filter down to real linkrevs
                        linkrevs = [l for l in linkrevs
                                    if self.lrugetctx(l)[f].filenode() == node]
                    except Exception:
                        pass
                self._warn(_(" (expected %s)") % " ".join
                           (map(pycompat.bytestr, linkrevs)))
            lr = None # can't be trusted

        try:
            p1, p2 = obj.parents(node)
            if p1 not in seen and p1 != nullid:
                self._err(lr, _("unknown parent 1 %s of %s") %
                    (short(p1), short(node)), f)
            if p2 not in seen and p2 != nullid:
                self._err(lr, _("unknown parent 2 %s of %s") %
                    (short(p2), short(node)), f)
        except Exception as inst:
            self._exc(lr, _("checking parents of %s") % short(node), inst, f)

        if node in seen:
            self._err(lr, _("duplicate revision %d (%d)") % (i, seen[node]), f)
        seen[node] = i
        return lr

    def verify(self):
        """verify the content of the Mercurial repository

        This method run all verifications, displaying issues as they are found.

        return 1 if any error have been encountered, 0 otherwise."""
        # initial validation and generic report
        repo = self.repo
        ui = repo.ui
        if not repo.url().startswith('file:'):
            raise error.Abort(_("cannot verify bundle or remote repos"))

        if os.path.exists(repo.sjoin("journal")):
            ui.warn(_("abandoned transaction found - run hg recover\n"))

        if ui.verbose or not self.revlogv1:
            ui.status(_("repository uses revlog format %d\n") %
                           (self.revlogv1 and 1 or 0))

        # data verification
        mflinkrevs, filelinkrevs = self._verifychangelog()
        filenodes = self._verifymanifest(mflinkrevs)
        del mflinkrevs
        self._crosscheckfiles(filelinkrevs, filenodes)
        totalfiles, filerevisions = self._verifyfiles(filenodes, filelinkrevs)

        # final report
        ui.status(_("checked %d changesets with %d changes to %d files\n") %
                       (len(repo.changelog), filerevisions, totalfiles))
        if self.warnings:
            ui.warn(_("%d warnings encountered!\n") % self.warnings)
        if self.fncachewarned:
            ui.warn(_('hint: run "hg debugrebuildfncache" to recover from '
                      'corrupt fncache\n'))
        if self.errors:
            ui.warn(_("%d integrity errors encountered!\n") % self.errors)
            if self.badrevs:
                ui.warn(_("(first damaged changeset appears to be %d)\n")
                        % min(self.badrevs))
            return 1
        return 0

    def _verifychangelog(self):
        """verify the changelog of a repository

        The following checks are performed:
        - all of `_checkrevlog` checks,
        - all of `_checkentry` checks (for each revisions),
        - each revision can be read.

        The function returns some of the data observed in the changesets as a
        (mflinkrevs, filelinkrevs) tuples:
        - mflinkrevs:   is a { manifest-node -> [changelog-rev] } mapping
        - filelinkrevs: is a { file-path -> [changelog-rev] } mapping

        If a matcher was specified, filelinkrevs will only contains matched
        files.
        """
        ui = self.ui
        repo = self.repo
        match = self.match
        cl = repo.changelog

        ui.status(_("checking changesets\n"))
        mflinkrevs = {}
        filelinkrevs = {}
        seen = {}
        self._checkrevlog(cl, "changelog", 0)
        progress = ui.makeprogress(_('checking'), unit=_('changesets'),
                                   total=len(repo))
        for i in repo:
            progress.update(i)
            n = cl.node(i)
            self._checkentry(cl, i, n, seen, [i], "changelog")

            try:
                changes = cl.read(n)
                if changes[0] != nullid:
                    mflinkrevs.setdefault(changes[0], []).append(i)
                    self.refersmf = True
                for f in changes[3]:
                    if match(f):
                        filelinkrevs.setdefault(_normpath(f), []).append(i)
            except Exception as inst:
                self.refersmf = True
                self._exc(i, _("unpacking changeset %s") % short(n), inst)
        progress.complete()
        return mflinkrevs, filelinkrevs

    def _verifymanifest(self, mflinkrevs, dir="", storefiles=None,
                        subdirprogress=None):
        """verify the manifestlog content

        Inputs:
        - mflinkrevs:     a {manifest-node -> [changelog-revisions]} mapping
        - dir:            a subdirectory to check (for tree manifest repo)
        - storefiles:     set of currently "orphan" files.
        - subdirprogress: a progress object

        This function checks:
        * all of `_checkrevlog` checks (for all manifest related revlogs)
        * all of `_checkentry` checks (for all manifest related revisions)
        * nodes for subdirectory exists in the sub-directory manifest
        * each manifest entries have a file path
        * each manifest node refered in mflinkrevs exist in the manifest log

        If tree manifest is in use and a matchers is specified, only the
        sub-directories matching it will be verified.

        return a two level mapping:
            {"path" -> { filenode -> changelog-revision}}

        This mapping primarily contains entries for every files in the
        repository. In addition, when tree-manifest is used, it also contains
        sub-directory entries.

        If a matcher is provided, only matching paths will be included.
        """
        repo = self.repo
        ui = self.ui
        match = self.match
        mfl = self.repo.manifestlog
        mf = mfl.getstorage(dir)

        if not dir:
            self.ui.status(_("checking manifests\n"))

        filenodes = {}
        subdirnodes = {}
        seen = {}
        label = "manifest"
        if dir:
            label = dir
            revlogfiles = mf.files()
            storefiles.difference_update(revlogfiles)
            if subdirprogress: # should be true since we're in a subdirectory
                subdirprogress.increment()
        if self.refersmf:
            # Do not check manifest if there are only changelog entries with
            # null manifests.
            self._checkrevlog(mf, label, 0)
        progress = ui.makeprogress(_('checking'), unit=_('manifests'),
                                   total=len(mf))
        for i in mf:
            if not dir:
                progress.update(i)
            n = mf.node(i)
            lr = self._checkentry(mf, i, n, seen, mflinkrevs.get(n, []), label)
            if n in mflinkrevs:
                del mflinkrevs[n]
            elif dir:
                self._err(lr, _("%s not in parent-directory manifest") %
                         short(n), label)
            else:
                self._err(lr, _("%s not in changesets") % short(n), label)

            try:
                mfdelta = mfl.get(dir, n).readdelta(shallow=True)
                for f, fn, fl in mfdelta.iterentries():
                    if not f:
                        self._err(lr, _("entry without name in manifest"))
                    elif f == "/dev/null":  # ignore this in very old repos
                        continue
                    fullpath = dir + _normpath(f)
                    if fl == 't':
                        if not match.visitdir(fullpath):
                            continue
                        subdirnodes.setdefault(fullpath + '/', {}).setdefault(
                            fn, []).append(lr)
                    else:
                        if not match(fullpath):
                            continue
                        filenodes.setdefault(fullpath, {}).setdefault(fn, lr)
            except Exception as inst:
                self._exc(lr, _("reading delta %s") % short(n), inst, label)
            if self._level >= VERIFY_FULL:
                try:
                    # Various issues can affect manifest. So we read each full
                    # text from storage. This triggers the checks from the core
                    # code (eg: hash verification, filename are ordered, etc.)
                    mfdelta = mfl.get(dir, n).read()
                except Exception as inst:
                    self._exc(lr, _("reading full manifest %s") % short(n),
                              inst, label)

        if not dir:
            progress.complete()

        if self.havemf:
            # since we delete entry in `mflinkrevs` during iteration, any
            # remaining entries are "missing". We need to issue errors for them.
            changesetpairs = [(c, m) for m in mflinkrevs for c in mflinkrevs[m]]
            for c, m in sorted(changesetpairs):
                if dir:
                    self._err(c, _("parent-directory manifest refers to unknown"
                                   " revision %s") % short(m), label)
                else:
                    self._err(c, _("changeset refers to unknown revision %s") %
                              short(m), label)

        if not dir and subdirnodes:
            self.ui.status(_("checking directory manifests\n"))
            storefiles = set()
            subdirs = set()
            revlogv1 = self.revlogv1
            for f, f2, size in repo.store.datafiles():
                if not f:
                    self._err(None, _("cannot decode filename '%s'") % f2)
                elif (size > 0 or not revlogv1) and f.startswith('meta/'):
                    storefiles.add(_normpath(f))
                    subdirs.add(os.path.dirname(f))
            subdirprogress = ui.makeprogress(_('checking'), unit=_('manifests'),
                                             total=len(subdirs))

        for subdir, linkrevs in subdirnodes.iteritems():
            subdirfilenodes = self._verifymanifest(linkrevs, subdir, storefiles,
                                                   subdirprogress)
            for f, onefilenodes in subdirfilenodes.iteritems():
                filenodes.setdefault(f, {}).update(onefilenodes)

        if not dir and subdirnodes:
            subdirprogress.complete()
            if self.warnorphanstorefiles:
                for f in sorted(storefiles):
                    self._warn(_("warning: orphan data file '%s'") % f)

        return filenodes

    def _crosscheckfiles(self, filelinkrevs, filenodes):
        repo = self.repo
        ui = self.ui
        ui.status(_("crosschecking files in changesets and manifests\n"))

        total = len(filelinkrevs) + len(filenodes)
        progress = ui.makeprogress(_('crosschecking'), unit=_('files'),
                                   total=total)
        if self.havemf:
            for f in sorted(filelinkrevs):
                progress.increment()
                if f not in filenodes:
                    lr = filelinkrevs[f][0]
                    self._err(lr, _("in changeset but not in manifest"), f)

        if self.havecl:
            for f in sorted(filenodes):
                progress.increment()
                if f not in filelinkrevs:
                    try:
                        fl = repo.file(f)
                        lr = min([fl.linkrev(fl.rev(n)) for n in filenodes[f]])
                    except Exception:
                        lr = None
                    self._err(lr, _("in manifest but not in changeset"), f)

        progress.complete()

    def _verifyfiles(self, filenodes, filelinkrevs):
        repo = self.repo
        ui = self.ui
        lrugetctx = self.lrugetctx
        revlogv1 = self.revlogv1
        havemf = self.havemf
        ui.status(_("checking files\n"))

        storefiles = set()
        for f, f2, size in repo.store.datafiles():
            if not f:
                self._err(None, _("cannot decode filename '%s'") % f2)
            elif (size > 0 or not revlogv1) and f.startswith('data/'):
                storefiles.add(_normpath(f))

        state = {
            # TODO this assumes revlog storage for changelog.
            'expectedversion': self.repo.changelog.version & 0xFFFF,
            'skipflags': self.skipflags,
            # experimental config: censor.policy
            'erroroncensored': ui.config('censor', 'policy') == 'abort',
        }

        files = sorted(set(filenodes) | set(filelinkrevs))
        revisions = 0
        progress = ui.makeprogress(_('checking'), unit=_('files'),
                                   total=len(files))
        for i, f in enumerate(files):
            progress.update(i, item=f)
            try:
                linkrevs = filelinkrevs[f]
            except KeyError:
                # in manifest but not in changelog
                linkrevs = []

            if linkrevs:
                lr = linkrevs[0]
            else:
                lr = None

            try:
                fl = repo.file(f)
            except error.StorageError as e:
                self._err(lr, _("broken revlog! (%s)") % e, f)
                continue

            for ff in fl.files():
                try:
                    storefiles.remove(ff)
                except KeyError:
                    if self.warnorphanstorefiles:
                        self._warn(_(" warning: revlog '%s' not in fncache!") %
                                  ff)
                        self.fncachewarned = True

            if not len(fl) and (self.havecl or self.havemf):
                self._err(lr, _("empty or missing %s") % f)
            else:
                # Guard against implementations not setting this.
                state['skipread'] = set()
                for problem in fl.verifyintegrity(state):
                    if problem.node is not None:
                        linkrev = fl.linkrev(fl.rev(problem.node))
                    else:
                        linkrev = None

                    if problem.warning:
                        self._warn(problem.warning)
                    elif problem.error:
                        self._err(linkrev if linkrev is not None else lr,
                                  problem.error, f)
                    else:
                        raise error.ProgrammingError(
                            'problem instance does not set warning or error '
                            'attribute: %s' % problem.msg)

            seen = {}
            for i in fl:
                revisions += 1
                n = fl.node(i)
                lr = self._checkentry(fl, i, n, seen, linkrevs, f)
                if f in filenodes:
                    if havemf and n not in filenodes[f]:
                        self._err(lr, _("%s not in manifests") % (short(n)), f)
                    else:
                        del filenodes[f][n]

                if n in state['skipread']:
                    continue

                # check renames
                try:
                    # This requires resolving fulltext (at least on revlogs). We
                    # may want ``verifyintegrity()`` to pass a set of nodes with
                    # rename metadata as an optimization.
                    rp = fl.renamed(n)
                    if rp:
                        if lr is not None and ui.verbose:
                            ctx = lrugetctx(lr)
                            if not any(rp[0] in pctx for pctx in ctx.parents()):
                                self._warn(_("warning: copy source of '%s' not"
                                            " in parents of %s") % (f, ctx))
                        fl2 = repo.file(rp[0])
                        if not len(fl2):
                            self._err(lr,
                                      _("empty or missing copy source revlog "
                                        "%s:%s") % (rp[0],
                                      short(rp[1])),
                                      f)
                        elif rp[1] == nullid:
                            ui.note(_("warning: %s@%s: copy source"
                                      " revision is nullid %s:%s\n")
                                % (f, lr, rp[0], short(rp[1])))
                        else:
                            fl2.rev(rp[1])
                except Exception as inst:
                    self._exc(lr, _("checking rename of %s") % short(n),
                              inst, f)

            # cross-check
            if f in filenodes:
                fns = [(v, k) for k, v in filenodes[f].iteritems()]
                for lr, node in sorted(fns):
                    self._err(lr, _("manifest refers to unknown revision %s") %
                              short(node), f)
        progress.complete()

        if self.warnorphanstorefiles:
            for f in sorted(storefiles):
                self._warn(_("warning: orphan data file '%s'") % f)

        return len(files), revisions