diff mercurial/exchangev2.py @ 39638:d292328e0143

exchangev2: fetch manifest revisions

Now that the server has support for retrieving manifest data, we can
implement the client bits to call it.

We teach the changeset fetching code to capture the manifest revisions
that are encountered on incoming changesets. We then feed this into a
new function which filters out known manifests and then batches up
manifest data requests to the server.

This is different from the previous wire protocol in a few notable
ways.

First, the client fetches manifest data separately and explicitly.
Before, we'd ask the server for data pertaining to some changesets
(via a "getbundle" command) and manifests (and files) would be sent
automatically. Providing an API for looking up just manifest data
separately gives clients much more flexibility for manifest management.
For example, a client may choose to only fetch manifest data on demand
instead of prefetching it (i.e. partial clone).

Second, we send N commands to the server for manifest retrieval instead
of 1. This property has a few nice side-effects. One is that the
deterministic nature of the requests lends itself to server-side
caching. For example, say the remote has 50,000 manifests. If the
server is configured to cache responses, each time a new commit
arrives, you will have a cache miss and need to regenerate all outgoing
data. But if you makes N requests requesting 10,000 manifests each,
a new commit will still yield cache hits on the initial, unchanged
manifest batches/requests.

A derived benefit from these properties is that resumable clone is
conceptually simpler to implement. When making a monolithic request
for all of the repository data, recovering from an interrupted clone
is hard because the server was in the driver's seat and was maintaining
state about all the data that needed transferred. With the client
driving fetching, the client can persist the set of unfetched entities
and retry/resume a fetch if something goes wrong. Or we can fetch all
data N changesets at a time and slowly build up a repository. This
approach is drastically easier to implement when we have server APIs
exposing low-level repository primitives (such as manifests and files).

We don't yet support tree manifests. But it should be possible to
implement that with the existing wire protocol command.

Differential Revision: https://phab.mercurial-scm.org/D4489

--- a/mercurial/exchangev2.py	Wed Sep 05 09:09:52 2018 -0700
+++ b/mercurial/exchangev2.py	Wed Sep 05 09:09:57 2018 -0700
@@ -16,6 +16,7 @@
 )
 from . import (
     bookmarks,
+    error,
     mdiff,
     phases,
     pycompat,
@@ -57,6 +58,8 @@
                                remote.url(), pullop.gettransaction,
                                explicit=pullop.explicitbookmarks)
 
+    _fetchmanifests(repo, tr, remote, csetres['manifestnodes'])
+
 def _pullchangesetdiscovery(repo, remote, heads, abortwhenunrelated=True):
     """Determine which changesets need to be pulled."""
 
@@ -121,6 +124,8 @@
                                     unit=_('chunks'),
                                     total=meta.get(b'totalitems'))
 
+    manifestnodes = {}
+
     def linkrev(node):
         repo.ui.debug('add changeset %s\n' % short(node))
         # Linkrev for changelog is always self.
@@ -129,6 +134,12 @@
     def onchangeset(cl, node):
         progress.increment()
 
+        revision = cl.changelogrevision(node)
+
+        # We need to preserve the mapping of changelog revision to node
+        # so we can set the linkrev accordingly when manifests are added.
+        manifestnodes[cl.rev(node)] = revision.manifest
+
     nodesbyphase = {phase: set() for phase in phases.phasenames}
     remotebookmarks = {}
 
@@ -178,4 +189,106 @@
         'added': added,
         'nodesbyphase': nodesbyphase,
         'bookmarks': remotebookmarks,
+        'manifestnodes': manifestnodes,
     }
+
+def _fetchmanifests(repo, tr, remote, manifestnodes):
+    rootmanifest = repo.manifestlog.getstorage(b'')
+
+    # Some manifests can be shared between changesets. Filter out revisions
+    # we already know about.
+    fetchnodes = []
+    linkrevs = {}
+    seen = set()
+
+    for clrev, node in sorted(manifestnodes.iteritems()):
+        if node in seen:
+            continue
+
+        try:
+            rootmanifest.rev(node)
+        except error.LookupError:
+            fetchnodes.append(node)
+            linkrevs[node] = clrev
+
+        seen.add(node)
+
+    # TODO handle tree manifests
+
+    # addgroup() expects 7-tuple describing revisions. This normalizes
+    # the wire data to that format.
+    def iterrevisions(objs, progress):
+        for manifest in objs:
+            node = manifest[b'node']
+
+            if b'deltasize' in manifest:
+                basenode = manifest[b'deltabasenode']
+                delta = next(objs)
+            elif b'revisionsize' in manifest:
+                basenode = nullid
+                revision = next(objs)
+                delta = mdiff.trivialdiffheader(len(revision)) + revision
+            else:
+                continue
+
+            yield (
+                node,
+                manifest[b'parents'][0],
+                manifest[b'parents'][1],
+                # The value passed in is passed to the lookup function passed
+                # to addgroup(). We already have a map of manifest node to
+                # changelog revision number. So we just pass in the
+                # manifest node here and use linkrevs.__getitem__ as the
+                # resolution function.
+                node,
+                basenode,
+                delta,
+                # Flags not yet supported.
+                0
+            )
+
+            progress.increment()
+
+    progress = repo.ui.makeprogress(_('manifests'), unit=_('chunks'),
+                                    total=len(fetchnodes))
+
+    # Fetch manifests 10,000 per command.
+    # TODO have server advertise preferences?
+    # TODO make size configurable on client?
+    batchsize = 10000
+
+    # We send commands 1 at a time to the remote. This is not the most
+    # efficient because we incur a round trip at the end of each batch.
+    # However, the existing frame-based reactor keeps consuming server
+    # data in the background. And this results in response data buffering
+    # in memory. This can consume gigabytes of memory.
+    # TODO send multiple commands in a request once background buffering
+    # issues are resolved.
+
+    added = []
+
+    for i in pycompat.xrange(0, len(fetchnodes), batchsize):
+        batch = [node for node in fetchnodes[i:i + batchsize]]
+        if not batch:
+            continue
+
+        with remote.commandexecutor() as e:
+            objs = e.callcommand(b'manifestdata', {
+                b'tree': b'',
+                b'nodes': batch,
+                b'fields': {b'parents', b'revision'},
+            }).result()
+
+            # Chomp off header object.
+            next(objs)
+
+            added.extend(rootmanifest.addgroup(
+                iterrevisions(objs, progress),
+                linkrevs.__getitem__,
+                weakref.proxy(tr)))
+
+    progress.complete()
+
+    return {
+        'added': added,
+    }