Patchwork [05,of,14] obsstore: add a method to incrementally retrieve obsmarkers

# HG changeset patch
# User Boris Feld <boris.feld@octobus.net>
# Date 1497674383 -7200
#      Sat Jun 17 06:39:43 2017 +0200
# Node ID 7c41d1484f53c53d516153d19a90c331c3f87c16
# Parent  e73d330f6fdb0cf1635920f4302b61b321ec58e2
# EXP-Topic obs-cache
obsstore: add a method to incrementally retrieve obsmarkers

Parsing the full obsstore is slow, so cache that depends on obsstore content
wants a way to know if the obsstore changed, and it this change was append
only.

For this purpose we introduce an official cache-key for the obsstore. This
cache-key work in a way similar to the '(tiprev, tipnode)' pair used for the
changelog. We use the size of the obsstore file and the hash of its tail. That
way, we can check if the obsstore grew and if the content we knew is still
present in the obsstore.

The cachekey comes with a method that only returns the obsmarkers that changed
between the last seen "cache-key" value and the current state of the
repository. That method is race free and should be used by caches. See
documentation for details.

note: we are reading the full obsstore file from disk as part of the key
validation process. This could be avoided but we keep the implementation
simple for now. Once it is in place and running we can iterate to make it
better.

On Sun, Jul 09, 2017 at 07:55:17PM +0200, Boris Feld wrote:
> # HG changeset patch
> # User Boris Feld <boris.feld@octobus.net>
> # Date 1497674383 -7200
> #      Sat Jun 17 06:39:43 2017 +0200
> # Node ID 7c41d1484f53c53d516153d19a90c331c3f87c16
> # Parent  e73d330f6fdb0cf1635920f4302b61b321ec58e2
> # EXP-Topic obs-cache
> obsstore: add a method to incrementally retrieve obsmarkers

This one and its follow-up both look interesting, but I don't know if
we have any imminent callers?

>
> Parsing the full obsstore is slow, so cache that depends on obsstore content
> wants a way to know if the obsstore changed, and it this change was append
> only.
>
> For this purpose we introduce an official cache-key for the obsstore. This
> cache-key work in a way similar to the '(tiprev, tipnode)' pair used for the
> changelog. We use the size of the obsstore file and the hash of its tail. That
> way, we can check if the obsstore grew and if the content we knew is still
> present in the obsstore.
>
> The cachekey comes with a method that only returns the obsmarkers that changed
> between the last seen "cache-key" value and the current state of the
> repository. That method is race free and should be used by caches. See
> documentation for details.
>
> note: we are reading the full obsstore file from disk as part of the key
> validation process. This could be avoided but we keep the implementation
> simple for now. Once it is in place and running we can iterate to make it
> better.
>
> diff -r e73d330f6fdb -r 7c41d1484f53 mercurial/obsolete.py
> --- a/mercurial/obsolete.py	Sun Jun 04 10:02:09 2017 -0700
> +++ b/mercurial/obsolete.py	Sat Jun 17 06:39:43 2017 +0200
> @@ -70,6 +70,7 @@
>  from __future__ import absolute_import
>
>  import errno
> +import hashlib
>  import struct
>
>  from .i18n import _
> @@ -513,6 +514,10 @@
>      # parents: (tuple of nodeid) or None, parents of precursors
>      #          None is used when no data has been recorded
>
> +    # how much data to read at the end of file,
> +    # 1024 byte should be about 10 markers in average
> +    _obskeyspan = 1024
> +
>      def __init__(self, svfs, defaultformat=_fm1version, readonly=False):
>          # caches for various obsolescence related cache
>          self.caches = {}
> @@ -540,6 +545,72 @@
>
>      __bool__ = __nonzero__
>
> +    def getmarkerssince(self, previouscontentkey):
> +        """retrieve all new markers (since "contentkey") + updated content key
> +
> +        This function is to be used by cache depending on obsstore content. It
> +        make sure cache can incrementally update themselves without fear
> +        obsstore stripping or race condition.  If the content key is invalid
> +        (some obsstore content got stripped), all markers in the obsstore will
> +        be returned.
> +
> +        return: (reset, obsmarkers, contentkey)
> +
> +        :reset: boolean, True if previouscontentkey was invalided. Previously
> +                stored data are invalid and should be discarded. Full markers
> +                content is return in this case.
> +
> +        :obsmarkers: the list of new obsmarkers.
> +
> +        :contentkey: a key matching the content of 'previouscontentkey' +
> +                     'obsmarkers'.
> +
> +        The content key is a pair of:
> +
> +            (obsstore size, hash of last N bytes of the obsstore)
> +
> +        It must be kept around by cache and provided again for the next
> +        incremental read of new obsmarkers.
> +        """
> +        # XXX This function could avoid loading the whole data from disk (and
> +        # only read new markers). It currently use 'self._data' to keep the
> +        # code simpler.
> +        keysize, keyhash = previouscontentkey
> +        fulldata = self._data
> +
> +        # check the existing cache key
> +        reset = False
> +        if len(fulldata) < keysize: # less data than expected, this is a strip
> +            reset = True
> +        else:
> +            if keysize == 0: # no obsstore
> +                actualhash = node.nullid
> +            else:
> +                first = max(0, keysize - self._obskeyspan)
> +                keydata = fulldata[first:keysize]
> +                actualhash = hashlib.sha1(keydata).digest()
> +            reset = actualhash != keyhash
> +        newsize = len(fulldata)
> +
> +        # read the new data
> +        if reset:
> +            keysize = None # read all data
> +        elif keysize == newsize:
> +            # no update since last change, exist early
> +            return False, [], previouscontentkey
> +        if newsize:
> +            start = max(0, newsize - self._obskeyspan)
> +            newhash = hashlib.sha1(fulldata[start:newsize]).digest()
> +            __, obsmarkers = _readmarkers(fulldata, keysize, newsize)
> +        else:
> +            # obsstore is empty, use a generic hash and skip reading markers.
> +            newhash == node.nullid
> +            obsmarkers = []
> +
> +        # for now and for the sake of simplicity make sure obsmarkers is a list
> +        obsmarkers = list(obsmarkers)
> +        return reset, obsmarkers, (newsize, newhash)
> +
>      @property
>      def readonly(self):
>          """True if marker creation is disabled
> _______________________________________________
> Mercurial-devel mailing list
> Mercurial-devel@mercurial-scm.org
> https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel