* [PATCH 00/82] public-inbox-imapd: read-only IMAP server
@ 2020-06-10 7:03 5% Eric Wong
2020-06-10 7:04 7% ` [PATCH 35/82] git: async: automatic retry on alternates change Eric Wong
0 siblings, 1 reply; 2+ results
From: Eric Wong @ 2020-06-10 7:03 UTC (permalink / raw)
To: meta
So I finally wrote my first IMAP server! And I'm actually
fairly satisfied with how it's turning out to support a bunch
of other performance + scalability work I've wanted to do.
Some previous notes here:
https://public-inbox.org/meta/20200609113442.GA16856@dcvr/
I finally seem to have gotten it to play nicely with mutt header
caching, so it's fit for public consumption :)
imaps://news.public-inbox.org/INBOX.comp.mail.public-inbox.meta.0
You can use any username+password, and AUTH=ANONYMOUS also
works if your client does that.
It doesn't support UTF-7 (mailbox names) or advertise UTF-8
in CAPABILITIES, yet; I still have RFCs to read :P
And there's a bunch of new things which could use some
testing from non-mutt/mbsync/offlineimap users.
Maybe you'll find some client-side bugs like I did :P
v1 reindexing also gets a little bit of parallelism :)
Anyways, I'll probably be porting some of the scalability
and slow-storage work to older parts of the code before
fiddling with more IMAP extensions.
Eric Wong (82):
doc: add some IMAP standards
nntpd: restrict allowed newsgroup names
preliminary imap server implementation
inboxidle: new class to detect inbox changes
imap: support IDLE
msgmap: split ->max into its own method
imap: delay InboxIdle start, support refresh
imap: implement STATUS command
imap: use Text::ParseWords::parse_line to handle quoted words
imap: support LIST command
t/imapd: support FakeInotify and KQNotify
imap: support fetch for BODYSTRUCTURE and BODY
eml: each_part: single part $idx is 1
imap: allow fetch of partial of BODY[...] and headers
imap: always include `resp-text' in responses
imap: split out unit tests and benchmarks
imap: fix multi-message partial header fetches
imap: simplify partial fetch structure
imap: support sequence number FETCH
imap: do not include ".PEEK" in responses
imap: support the CLOSE command
imap: speed up HEADER.FIELDS[.NOT] range fetches
git: async: flatten the inflight array
git: do our own read buffering for cat-file
imap: use git-cat-file asynchronously
git: idle rbuf for async
imap: support LSUB command
imap: FETCH: support comma-delimited ranges
add imapd compression test
testcommon: tcp_(server|connect): BAIL_OUT on failure
*deflate: drop invalid comment about rbuf
imap: fix pipelining with async git
git: cat_async: provide requested OID + "missing" on missing blobs
git: move async_cat reference to PublicInbox::Git
git: async: automatic retry on alternates change
imapclient: wrapper for Mail::IMAPClient
xt: add imapd-validate and imapd-mbsync-oimap
imap: support out-of-bounds ranges
xt/perf-imap-list: time refresh_inboxlist
imap: case-insensitive mailbox name comparisons
imap: break giant inboxes into sub-inboxes of 50K messages
imap: start introducing iterative config reloading
imap: require ".$UID_MIN-$UID_END" suffix
imapd: ensure LIST is sorted alphabetically, for now
imap: omit $UID_END from mailbox name, use index
t/config.t: always compare against git bool behavior
xt/*: show some tunable parameters
imap: STATUS and LIST are case-insensitive, too
imap: EXAMINE/STATUS: return correct counts
imap: avoid uninitialized warnings on incomplete commands
imap: start parsing out queries for SQLite and Xapian
imap: SEARCH: clamp results to the 50K UID range
imap: allow UID range search on timestamps
over: get_art: use dbh->prepare_cached
search: index byte size of a message for IMAP search
search: index UID for IMAP search, too
imap: remove dummies from sequence number FETCH
imap: compile UID FETCH to opcodes
imap: UID FETCH: optimize for smsg-only case
imap: UID FETCH: optimize (UID FLAGS) harder
imap: IDLE: avoid extraneous wakeups, keep-alive
imap: 30 minute auto-logout timer
imap: split ->logged_in attribute into a separate class
searchidx: v1 (re)-index uses git asynchronously
index: account for CRLF conversion when storing bytes
imap: rely on smsg->{bytes} for RFC822.SIZE
imap: UID FETCH requires at least one data item
imap: LIST shows "INBOX" in all caps
imap: support 8000 octet lines
imap: reinstate some message sequence number support
imap: cleanup ->{uid_base} usage
imap: FETCH: more granular CRLF conversion
imap: further speed up HEADER.FIELDS FETCH requests
imap: FETCH: try to make fake MSNs sequentially
imap: STATUS/EXAMINE: rely on SQLite overview
imap: UID SEARCH: support multiple ranges
imap: wire up Xapian search, msn SEARCH and multiple ranges
imap: misc cleanups and notes
imapd: don't bother sorting LIST output
imap: drop non-UID SEARCH for now
over: uid_range: remove LIMIT
imap: FETCH: proper MSN => UID mapping for requests
Documentation/public-inbox-imapd.pod | 91 ++
Documentation/standards.perl | 10 +
MANIFEST | 18 +
lib/PublicInbox/Config.pm | 18 +
lib/PublicInbox/Daemon.pm | 24 +-
lib/PublicInbox/DummyInbox.pm | 22 +
lib/PublicInbox/Eml.pm | 9 +-
lib/PublicInbox/FakeInotify.pm | 59 ++
lib/PublicInbox/Git.pm | 163 +--
lib/PublicInbox/GitAsyncCat.pm | 51 +
lib/PublicInbox/IMAP.pm | 1397 ++++++++++++++++++++++++++
lib/PublicInbox/IMAPClient.pm | 119 +++
lib/PublicInbox/IMAPD.pm | 114 +++
lib/PublicInbox/IMAPdeflate.pm | 126 +++
lib/PublicInbox/Import.pm | 2 +-
lib/PublicInbox/In2Tie.pm | 17 +
lib/PublicInbox/Inbox.pm | 33 +-
lib/PublicInbox/InboxIdle.pm | 79 ++
lib/PublicInbox/KQNotify.pm | 66 ++
lib/PublicInbox/Lock.pm | 7 +
lib/PublicInbox/MsgIter.pm | 2 +-
lib/PublicInbox/Msgmap.pm | 20 +-
lib/PublicInbox/NNTPD.pm | 12 +-
lib/PublicInbox/NNTPdeflate.pm | 1 -
lib/PublicInbox/Over.pm | 50 +-
lib/PublicInbox/Search.pm | 32 +-
lib/PublicInbox/SearchIdx.pm | 89 +-
lib/PublicInbox/SearchIdxShard.pm | 11 +-
lib/PublicInbox/Smsg.pm | 8 +-
lib/PublicInbox/TestCommon.pm | 7 +-
lib/PublicInbox/V2Writable.pm | 10 +-
script/public-inbox-imapd | 14 +
t/config.t | 15 +-
t/eml.t | 2 +-
t/git.t | 40 +-
t/imap.t | 133 +++
t/imapd-tls.t | 204 ++++
t/imapd.t | 398 ++++++++
t/import.t | 5 +-
t/inbox_idle.t | 72 ++
t/nntpd.t | 5 +-
t/over.t | 3 +
t/search.t | 19 +
xt/cmp-msgstr.t | 1 -
xt/cmp-msgview.t | 1 -
xt/eml_check_limits.t | 6 +-
xt/git_async_cmp.t | 2 +-
xt/imapd-mbsync-oimap.t | 132 +++
xt/imapd-validate.t | 177 ++++
xt/mem-msgview.t | 1 +
xt/msgtime_cmp.t | 1 -
xt/perf-msgview.t | 1 -
52 files changed, 3718 insertions(+), 181 deletions(-)
create mode 100644 Documentation/public-inbox-imapd.pod
create mode 100644 lib/PublicInbox/DummyInbox.pm
create mode 100644 lib/PublicInbox/FakeInotify.pm
create mode 100644 lib/PublicInbox/GitAsyncCat.pm
create mode 100644 lib/PublicInbox/IMAP.pm
create mode 100644 lib/PublicInbox/IMAPClient.pm
create mode 100644 lib/PublicInbox/IMAPD.pm
create mode 100644 lib/PublicInbox/IMAPdeflate.pm
create mode 100644 lib/PublicInbox/In2Tie.pm
create mode 100644 lib/PublicInbox/InboxIdle.pm
create mode 100644 lib/PublicInbox/KQNotify.pm
create mode 100644 script/public-inbox-imapd
create mode 100644 t/imap.t
create mode 100644 t/imapd-tls.t
create mode 100644 t/imapd.t
create mode 100644 t/inbox_idle.t
create mode 100644 xt/imapd-mbsync-oimap.t
create mode 100644 xt/imapd-validate.t
^ permalink raw reply [relevance 5%]
* [PATCH 35/82] git: async: automatic retry on alternates change
2020-06-10 7:03 5% [PATCH 00/82] public-inbox-imapd: read-only IMAP server Eric Wong
@ 2020-06-10 7:04 7% ` Eric Wong
0 siblings, 0 replies; 2+ results
From: Eric Wong @ 2020-06-10 7:04 UTC (permalink / raw)
To: meta
This matches the behavior of the existing synchronous ->cat_file
method. In fact, ->cat_file now becomes a small wrapper around
the ->cat_async method.
---
lib/PublicInbox/Git.pm | 64 +++++++++++++++++++++++++-----------------
t/git.t | 37 ++++++++++++++++++++----
2 files changed, 70 insertions(+), 31 deletions(-)
diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index 60236afe7bf..a55c48d5f52 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -18,6 +18,7 @@ use base qw(Exporter);
our @EXPORT_OK = qw(git_unquote git_quote);
use Errno qw(EINTR);
our $PIPE_BUFSIZ = 65536; # Linux default
+our $in_cleanup;
use constant MAX_INFLIGHT =>
(($^O eq 'linux' ? 4096 : POSIX::_POSIX_PIPE_BUF()) * 3)
@@ -155,6 +156,26 @@ sub my_readline ($$) {
}
}
+sub cat_async_retry ($$$$$) {
+ my ($self, $inflight, $req, $cb, $arg) = @_;
+
+ # {inflight} may be non-existent, but if it isn't we delete it
+ # here to prevent cleanup() from waiting:
+ delete $self->{inflight};
+ cleanup($self);
+
+ $self->{inflight} = $inflight;
+ batch_prepare($self);
+ my $buf = "$req\n";
+ for (my $i = 0; $i < @$inflight; $i += 3) {
+ $buf .= "$inflight->[$i]\n";
+ }
+ print { $self->{out} } $buf or fail($self, "write error: $!");
+ unshift(@$inflight, \$req, $cb, $arg); # \$ref to indicate retried
+
+ cat_async_step($self, $inflight); # take one step
+}
+
sub cat_async_step ($$) {
my ($self, $inflight) = @_;
die 'BUG: inflight empty or odd' if scalar(@$inflight) < 3;
@@ -168,8 +189,13 @@ sub cat_async_step ($$) {
fail($self, defined($bref) ? 'read EOF' : "read: $!");
chop($$bref) eq "\n" or fail($self, 'LF missing after blob');
} elsif ($head =~ / missing$/) {
+ # ref($req) indicates it's already been retried
+ if (!ref($req) && !$in_cleanup && alternates_changed($self)) {
+ return cat_async_retry($self, $inflight,
+ $req, $cb, $arg);
+ }
$type = 'missing';
- $oid = $req;
+ $oid = ref($req) ? $$req : $req;
} else {
fail($self, "Unexpected result from async git cat-file: $head");
}
@@ -190,33 +216,18 @@ sub batch_prepare ($) {
_bidi_pipe($_[0], qw(--batch in out pid));
}
+sub _cat_file_cb {
+ my ($bref, undef, undef, $size, $result) = @_;
+ @$result = ($bref, $size);
+}
+
sub cat_file {
- my ($self, $obj, $sizeref) = @_;
- my ($retried, $head, $rbuf);
+ my ($self, $oid, $sizeref) = @_;
+ my $result = [];
+ cat_async($self, $oid, \&_cat_file_cb, $result);
cat_async_wait($self);
-again:
- batch_prepare($self);
- $rbuf = delete($self->{cat_rbuf}) // \(my $new = '');
- print { $self->{out} } $obj, "\n" or fail($self, "write error: $!");
- $head = my_readline($self->{in}, $rbuf);
- if ($head =~ / missing$/) {
- if (!$retried && alternates_changed($self)) {
- $retried = 1;
- cleanup($self);
- goto again;
- }
- return;
- }
- $head =~ /^[0-9a-f]{40} \S+ ([0-9]+)$/ or
- fail($self, "Unexpected result from git cat-file: $head");
-
- my $size = $1 + 0;
- $$sizeref = $size if $sizeref;
- my $ret = my_read($self->{in}, $rbuf, $size + 1);
- $self->{cat_rbuf} = $rbuf if $$rbuf ne '';
- fail($self, defined($ret) ? 'read EOF' : "read: $!") if !$ret;
- chop($$ret) eq "\n" or fail($self, 'newline missing after blob');
- $ret;
+ $$sizeref = $result->[1] if $sizeref;
+ $result->[0];
}
sub check {
@@ -283,6 +294,7 @@ sub qx {
# returns true if there are pending "git cat-file" processes
sub cleanup {
my ($self) = @_;
+ local $in_cleanup = 1;
if (my $ac = $self->{async_cat}) {
$ac->close; # PublicInbox::GitAsyncCat::close -> EPOLL_CTL_DEL
}
diff --git a/t/git.t b/t/git.t
index 98d16f289c3..228df90fa45 100644
--- a/t/git.t
+++ b/t/git.t
@@ -86,23 +86,50 @@ if (1) {
if ('alternates reloaded') {
my ($alt, $alt_obj) = tmpdir();
- my @cmd = ('git', "--git-dir=$alt", qw(hash-object -w --stdin));
+ my $hash_obj = [ 'git', "--git-dir=$alt", qw(hash-object -w --stdin) ];
PublicInbox::Import::init_bare($alt);
open my $fh, '<', "$alt/config" or die "open failed: $!\n";
- my $rd = popen_rd(\@cmd, {}, { 0 => $fh } );
- close $fh or die "close failed: $!";
- chomp(my $remote = <$rd>);
+ chomp(my $remote = xqx($hash_obj, undef, { 0 => $fh }));
my $gcf = PublicInbox::Git->new($dir);
is($gcf->cat_file($remote), undef, "remote file not found");
open $fh, '>>', "$dir/objects/info/alternates" or
die "open failed: $!\n";
- print $fh "$alt/objects" or die "print failed: $!\n";
+ print $fh "$alt/objects\n" or die "print failed: $!\n";
close $fh or die "close failed: $!";
my $found = $gcf->cat_file($remote);
open $fh, '<', "$alt/config" or die "open failed: $!\n";
my $config = eval { local $/; <$fh> };
is($$found, $config, 'alternates reloaded');
+ # with the async interface
+ my ($async_alt, $async_dir_obj) = tmpdir();
+ PublicInbox::Import::init_bare($async_alt);
+ my @exist = map { chomp; [ split / / ] } (xqx(['git', "--git-dir=$dir",
+ qw(cat-file --batch-all-objects --batch-check)]));
+ my $results = [];
+ my $cb = sub {
+ my ($bref, $oid, $type, $size) = @_;
+ push @$results, [ $oid, $type, $size ];
+ };
+ for my $i (0..5) {
+ $gcf->cat_async($exist[$i]->[0], $cb, $results);
+ next if $i != 3;
+
+ # stick a new alternate into a running async pipeline
+ $hash_obj->[1] = "--git-dir=$async_alt";
+ $remote = xqx($hash_obj, undef, { 0 => \'async' });
+ chomp $remote;
+ open $fh, '>>', "$dir/objects/info/alternates" or
+ die "open failed: $!\n";
+ print $fh "$async_alt/objects\n" or die "print failed: $!\n";
+ close $fh or die "close failed: $!";
+ # trigger cat_async_retry:
+ $gcf->cat_async($remote, $cb, $results);
+ }
+ $gcf->cat_async_wait;
+ my $expect = [ @exist[0..3], [ $remote, 'blob', 5 ], @exist[4..5] ];
+ is_deeply($results, $expect, 'got expected results');
+
ok(!$gcf->cleanup, 'cleanup can expire');
ok(!$gcf->cleanup, 'cleanup idempotent');
^ permalink raw reply related [relevance 7%]
Results 1-2 of 2 | reverse | options above
-- pct% links below jump to the message on this page, permalinks otherwise --
2020-06-10 7:03 5% [PATCH 00/82] public-inbox-imapd: read-only IMAP server Eric Wong
2020-06-10 7:04 7% ` [PATCH 35/82] git: async: automatic retry on alternates change Eric Wong
Code repositories for project(s) associated with this public inbox
https://80x24.org/public-inbox.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).