* sp/maint-fetch-pack-stop-early:
enable "no-done" extension only when fetching over smart-http
* sp/maint-upload-pack-stop-early:
enable "no-done" extension only when serving over smart-http
Last night I had to make these two emergency reverts, but now we have a
better understanding of which part of the topic was broken, let's get rid
of the revert to fix it correctly.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The fetch-pack/upload-pack protocol relies on the underlying transport
(local pipe or TCP socket) to have enough slack to allow one window worth
of data in flight without blocking the writer. Traditionally we always
relied on being able to have two windows of 32 "have"s in flight (roughly
3k bytes) to stream.
The recent "progressive-stride" change allows "fetch-pack" to send up to
1024 "have"s without reading any response from "upload-pack". The
outgoing pipe of "upload-pack" can be clogged with many ACK and NAK that
are unread, while "fetch-pack" is still stuffing its outgoing pipe with
more "have"s, leading to a deadlock.
Revert the change unless we are in stateless rpc (aka smart-http) mode, as
using a large window full of "have"s is still a good way to help reduce
the number of back-and-forth, and there is no buffering issue there (it is
strictly "ping-pong" without an overlap).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 'no-done' protocol extension is used, the upload-pack (i.e. the
server side) process stops listening to the fetch-pack after issuing the
final NAK, and starts sending the generated pack data back, but there may
be more "have" send by the latter in flight that the fetch-pack is
expecting to be responded with ACK/NAK. This will typically result in a
deadlock (both will block on write that the other end never reads) or
SIGPIPE on the fetch-pack end (upload-pack will finish writing a small
pack and goes away).
Disable it unless fetch-pack is running under smart-http, where there is
no such streaming issue.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
* jc/fetch-progressive-stride:
fetch-pack: use smaller handshake window for initial request
fetch-pack: progressively use larger handshake windows
fetch-pack: factor out hardcoded handshake window size
Conflicts:
builtin/fetch-pack.c
* jc/maint-fetch-alt:
fetch-pack: objects in our alternates are available to us
refs_from_alternate: helper to use refs from alternates
Conflicts:
builtin/receive-pack.c
Fix warnings from 'make check'.
- These files don't include 'builtin.h' causing sparse to complain that
cmd_* isn't declared:
builtin/clone.c:364, builtin/fetch-pack.c:797,
builtin/fmt-merge-msg.c:34, builtin/hash-object.c:78,
builtin/merge-index.c:69, builtin/merge-recursive.c:22
builtin/merge-tree.c:341, builtin/mktag.c:156, builtin/notes.c:426
builtin/notes.c:822, builtin/pack-redundant.c:596,
builtin/pack-refs.c:10, builtin/patch-id.c:60, builtin/patch-id.c:149,
builtin/remote.c:1512, builtin/remote-ext.c:240,
builtin/remote-fd.c:53, builtin/reset.c:236, builtin/send-pack.c:384,
builtin/unpack-file.c:25, builtin/var.c:75
- These files have symbols which should be marked static since they're
only file scope:
submodule.c:12, diff.c:631, replace_object.c:92, submodule.c:13,
submodule.c:14, trace.c:78, transport.c:195, transport-helper.c:79,
unpack-trees.c:19, url.c:3, url.c:18, url.c:104, url.c:117, url.c:123,
url.c:129, url.c:136, thread-utils.c:21, thread-utils.c:48
- These files redeclare symbols to be different types:
builtin/index-pack.c:210, parse-options.c:564, parse-options.c:571,
usage.c:49, usage.c:58, usage.c:63, usage.c:72
- These files use a literal integer 0 when they really should use a NULL
pointer:
daemon.c:663, fast-import.c:2942, imap-send.c:1072, notes-merge.c:362
While we're in the area, clean up some unused #includes in builtin files
(mostly exec_cmd.h).
Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Start the initial request small by halving the INITIAL_FLUSH (we will try
to stay one window ahead of the server, so we would end up giving twice as
many "have" in flight at the very beginning). We may want to tweak these
values even more, taking MTU into account.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn Pearce <spearce@spearce.org>
The client has to dig the history deeper when more recent parts of its
history do not have any overlap with the server it is fetching from. Make
the handshake window exponentially larger as we dig deeper, with a
reasonable upper cap.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn Pearce <spearce@spearce.org>
The "git fetch" client presents the most recent 32 commits it has to the
server and gives a chance to the server to say "ok, we heard enough", and
continues reporting what it has in chunks of 32 commits, digging its
history down to older commits.
Move the hardcoded size of the handshake window outside the code, so that
we can tweak it more easily.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn Pearce <spearce@spearce.org>
Use the helper function split from the receiving end of "git push" to
allow the same optimization on the receiving end of "git fetch".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
If enabled on the connection "multi_ack_detailed no-done" as a
pair allows the remote upload-pack process to send a PACK down
to the client as soon as a "ACK %s ready" message was also sent.
Over git:// and ssh:// where a bi-directional stream is in place
this has very little difference over the classical version that
waits for the client to send a "done\n" line by itself. It does
slightly reduce the latency involved to start the pack stream as
there is one less round-trip from client->server required.
Over smart HTTP this avoids needing to send a final RPC that has
all of the prior common objects. Instead the server is able to
return a pack as soon as its ready to. For many common users the
smart HTTP fetch is now just 2 requests: GET .../info/refs, and
a POST .../git-upload-pack to not only negotiate but also receive
the pack stream. Only users who have more than 32 local unshared
commits with the remote will need additional requests to negotiate
a common merge base.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If multi_ack_detailed was selected in the protocol capabilities
(both client and server are >= Git 1.6.6) the upload-pack side will
send "ACK %s ready" when it knows how to safely cut the graph and
produce a reasonable pack for the want list that was already sent
on the connection.
Upon receiving "ACK %s ready" there is no point in looking at
the remaining commits inside of rev_list. Sending additional
"have %s" lines to the remote will not construct a smaller pack.
It is unlikely a commit older than the current cut point will have
a better delta base than the cut point itself has.
The original design of this code had fetch-pack empty rev_list by
marking a commit and its transitive ancestors COMMON whenever the
remote side said "ACK %s {continue,common}" and skipping over any
already COMMON commits during get_rev(). This approach does not
work when most of rev_list is actually COMMON_REF, commits that
are pointed to by a reference on the remote, which exist locally,
and which have not yet been sent to the remote as a "have %s" line.
Most of the common references are tags in the ref/tags namespace,
using points in the commit graph that are more than 1 commit apart.
In git.git itself, this is currently 340 tags, 339 of which point to
commits in the commit graph. fetch-pack pushes all of these into
rev_list, but is unable to mark them COMMON and discard during a
remote's "ACK %s {continue,common}" because it does not parse through
the entire parent chain. Not parsing the entire parent chain is
an optimization to avoid walking back to the roots of the repository.
Assuming the client is only following the remote (and does not make
its own local commits), the client needs 11 rounds to spin through
the entire list of tags (32 commits per round, ceil(339/32) == 11).
Unfortunately the server knows on the first "have %s" line that
it can produce a good pack, and does not need to see the remaining
320 tags in the other 10 rounds.
Over git:// and ssh:// this isn't as bad as it sounds, the client is
only transmitting an extra 16,000 bytes that it doesn't need to send.
Over smart HTTP, the client must do an additional 10 HTTP POST
requests, each of which incurs round-trip latency, and must upload
the entire state vector of all known common objects. On the final
POST request, this is 16 KiB worth of data.
Fix all of this by clearing rev_list as soon as the remote side
says it can construct a pack.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This shows a trace of all packets coming in or out of a given
program. This can help with debugging object negotiation or
other protocol issues.
To keep the code changes simple, we operate at the lowest
level, meaning we don't necessarily understand what's in the
packets. The one exception is a packet starting with "PACK",
which causes us to skip that packet and turn off tracing
(since the gigantic pack data will not be interesting to
read, at least not in the trace format).
We show both written and read packets. In the local case,
this may mean you will see packets twice (written by the
sender and read by the receiver). However, for cases where
the other end is remote, this allows you to see the full
conversation.
Packet tracing can be enabled with GIT_TRACE_PACKET=<foo>,
where <foo> takes the same arguments as GIT_TRACE.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add commit_list prefix to insert_by_date function and to sort_by_date,
so it's clear that these functions refer to commit_list structure.
Signed-off-by: Thiago Farina <tfransosi@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This shrinks the top-level directory a bit, and makes it much more
pleasant to use auto-completion on the thing. Instead of
[torvalds@nehalem git]$ em buil<tab>
Display all 180 possibilities? (y or n)
[torvalds@nehalem git]$ em builtin-sh
builtin-shortlog.c builtin-show-branch.c builtin-show-ref.c
builtin-shortlog.o builtin-show-branch.o builtin-show-ref.o
[torvalds@nehalem git]$ em builtin-shor<tab>
builtin-shortlog.c builtin-shortlog.o
[torvalds@nehalem git]$ em builtin-shortlog.c
you get
[torvalds@nehalem git]$ em buil<tab> [type]
builtin/ builtin.h
[torvalds@nehalem git]$ em builtin [auto-completes to]
[torvalds@nehalem git]$ em builtin/sh<tab> [type]
shortlog.c shortlog.o show-branch.c show-branch.o show-ref.c show-ref.o
[torvalds@nehalem git]$ em builtin/sho [auto-completes to]
[torvalds@nehalem git]$ em builtin/shor<tab> [type]
shortlog.c shortlog.o
[torvalds@nehalem git]$ em builtin/shortlog.c
which doesn't seem all that different, but not having that annoying
break in "Display all 180 possibilities?" is quite a relief.
NOTE! If you do this in a clean tree (no object files etc), or using an
editor that has auto-completion rules that ignores '*.o' files, you
won't see that annoying 'Display all 180 possibilities?' message - it
will just show the choices instead. I think bash has some cut-off
around 100 choices or something.
So the reason I see this is that I'm using an odd editory, and thus
don't have the rules to cut down on auto-completion. But you can
simulate that by using 'ls' instead, or something similar.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
