Eric Wong [Wed, 20 Jan 2021 05:04:43 +0000 (14:04 +0900)]
lei: allow more mbox inode types
We may attempt to write an mbox to any terminal, block, or
character device, not just regular files and FIFOs/pipes.
The only thing that is known to not work is a directory.
Sockets may be possible with some OSes (e.g. Plan 9) or
filesystems. This fixes t/lei.t on FreeBSD 11.x
Eric Wong [Tue, 19 Jan 2021 09:34:33 +0000 (09:34 +0000)]
t/lei: fix double-running of socket test with oneshot
We split out t/lei-oneshot.t and t/lei.t so it's easier
to isolate run-mode specific bugs and behavior and there's
no reason to rerun the socket daemon tests.
Eric Wong [Tue, 19 Jan 2021 09:34:31 +0000 (09:34 +0000)]
lei q: fix augment of compressed mailboxes
We need to delay writing out the mailbox until the compressor
process is up and running, so have startq wait a bit. This
means we must create the pipe early and hand it off to the
workers before augmenting, despite spawning the
gzip/pigz/xz/bzip2 process after augment is complete.
Eric Wong [Tue, 19 Jan 2021 09:34:30 +0000 (09:34 +0000)]
lei: write daemon errors to the sock directory
Most everything should be captured by the __WARN__ handlers and
routed to syslog, but it appears Perl may write to stderr in
some emergency cases, as can libc or other libraries. Just
point it to a small file that's cleared on reboot.
Eric Wong [Tue, 19 Jan 2021 09:34:28 +0000 (09:34 +0000)]
lei q: fix SIGPIPE handling from lei2mail workers
We need to properly propagate SIGPIPE to the top-level
lei-daemon process and avoid relying on auto-close,
since auto-close triggers Perl warnings when explicit
close() does not.
Eric Wong [Tue, 19 Jan 2021 09:34:27 +0000 (09:34 +0000)]
lei q: start ->mset while query_prepare runs
We don't need the result of query_prepare (for augmenting or
mass unlinking) until we're ready to deduplicate and write
results to the filesystem. This ought to let us hide some of
the cost of Xapian searches on multi-device/core systems for
extremely expensive searches.
Eric Wong [Mon, 18 Jan 2021 10:30:32 +0000 (04:30 -0600)]
lei_to_mail: optimize for MUAs
Instead of optimizing our own performance, this optimizes
our data to reduce work done by the MUA consumer.
Maildir and mbox destinations no longer support any notion of
the IMAP \Recent flag. JMAP has no functioning \Recent
equivalent, and neither do we.
In practice, having MUAs (e.g. mutt) clear the \Recent flag when
committing changes to the mbox is expensive: it creates a
rename(2) storm with Maildir and overwrites the entire mbox.
For mboxcl2 (and mboxcl), we'll further optimize mutt behavior
by setting the Lines: header in addition to Content-Length.
With these changes, mutt exits instantaneously on mboxcl2,
mboxcl, and Maildirs generated by "lei q".
Eric Wong [Mon, 18 Jan 2021 10:30:31 +0000 (04:30 -0600)]
lei q: parallelize Maildir and mbox writing
With 4 dedicated workers, this seems to provide a 100-120%
speedup on a 4 core machine when writing thousands of search
results to a Maildir or mbox. This also sets us up for
high-latency IMAP destinations in the future.
This opens the door to more speedup opportunities such
as optimizing dedupe locking and other ways to reduce
contention.
This change is fairly complex and convoluted, unfortunately.
Further work may allow us to simplify it and even improve
performance.
Eric Wong [Sun, 17 Jan 2021 08:52:27 +0000 (20:52 -1200)]
lei q: add --mua-cmd switch
It can be convenient to invoke an MUA as search results
are being written to it, as an eager person may want to
start seeing results ASAP. This lets Maildir users
see results in the MUA as we are writing them. Users
of IMAP will eventually be able to take advantage of
them, too.
Since we don't support mbox locking (yet?), we'll only invoke
the MUA after results are done for mbox formats.
Eric Wong [Sat, 16 Jan 2021 11:36:23 +0000 (23:36 -1200)]
lei: q: results output to Maildir and mbox* working
All the augment and deduplication stuff seems to be working
based on unit tests. OpPipe is a nice general addition that
will probably make future state machines easier.
Eric Wong [Sat, 16 Jan 2021 11:36:22 +0000 (23:36 -1200)]
ipc: children don't kill on DESTROY, reduce FD sharing
Children should not be blindly killing siblings on ->DESTROY
since they're typically shorter-lived than parents. We'll
also be more careful about on-stack variables and now we
can rely exclusively on delete ops to close FDs.
We also need to fix our SIGPIPE handling for the oneshot case
while fixing a typo for delete, so we write "!" to the EOF pipe
to ensure the parent oneshot process exits on the first worker
that hits SIGPIPE, rather than waiting for the last worker to
hit SIGPIPE.
Eric Wong [Sun, 17 Jan 2021 07:09:59 +0000 (07:09 +0000)]
extindex: fix w/ Xapian 1.2.21..1.2.24
Xapian v1.2.21..v1.2.24 failed to set the close-on-exec flag
on the flintlock FD, causing "git cat-file" processes to
hold onto the lock and prevent subsequent Xapian::WritableDatabase
from locking the DB. So cleanup git processes after committing
the miscidx transaction.
Eric Wong [Sun, 17 Jan 2021 07:09:58 +0000 (07:09 +0000)]
t/shared_kv: workaround old File::Spec
The version of File::Spec shipped with Perl 5.16.3 memoizes the
value of File::Spec->tmpdir, causing changes to $ENV{TMPDIR}
after-the-fact to be ignored.
We'll only work around this in the test since it's innocuous and
unlikely to matter in real-world usage (and there's many places
where we'd have to workaround this in non-test code).
Eric Wong [Sun, 17 Jan 2021 07:09:56 +0000 (07:09 +0000)]
initialize scalar for `vec' perlop modification
Older Perls (tested 5.16.3) would warn on uninitialized scalars while
newer (tested 5.28.1) do not. Just initialize it to an empty string
since it'll be filled in by `vec'.
Eric Wong [Thu, 14 Jan 2021 07:06:27 +0000 (19:06 -1200)]
lei: pass FD to CWD via cmsg, use fchdir on server
Perl chdir() automatically does fchdir(2) if given a file
or directory handle since 5.8.8/5.10.0, so we can safely
rely on it given our 5.10.1+ requirement.
This means we no longer have to waste several milliseconds
loading the Cwd.so and making stat() calls to ensure
ENV{PWD} is correct and usable in the server. It also lets
us work in directories that are no longer accessible via
pathname.
Eric Wong [Thu, 14 Jan 2021 07:06:24 +0000 (19:06 -1200)]
lei: q: lock stdout on overview output
Most writes to stdout aren't atomic and we need locking to
prevent workers from interleaving and corrupting JSON output.
The one case stdout won't require locking is if it's pointed
to a regular file with O_APPEND; as POSIX O_APPEND semantics
guarantees atomicity.
Eric Wong [Thu, 14 Jan 2021 07:06:23 +0000 (19:06 -1200)]
lei_overview: rename "references" to "refs"
"references" was too long of a name compared to the other field
names we output in the JSON. While we currently don't have a
"refs:" search prefix for the "References:" header, we may in
the future.
Eric Wong [Thu, 14 Jan 2021 07:06:22 +0000 (19:06 -1200)]
search: rename "ts:" prefix to "rt:"
Meaning "Received time", as it is the best description of the
value we use from the "Received:" header, if present. JMAP
calls it "receivedAt", but "rt:" seems like a better
abbreviation being in line with "dt:" for the "Date" header.
"Timestamp" ("ts") was potentially ambiguous given the presence
of the "Date" header.
Eric Wong [Thu, 14 Jan 2021 07:06:18 +0000 (19:06 -1200)]
lei: reduce live FD references in wq child
We can shrink the @TO_CLOSE_ATFORK_CHILD array by two
elements, at least. I may be possible to eliminate this
array entirely but clobbering $quit doesn't seem to
remove references to $eof_w or the $listener socket.
Eric Wong [Thu, 14 Jan 2021 07:06:17 +0000 (19:06 -1200)]
lei: do not unlink socket path at exit
This matches existing -httpd/-nntpd/-imapd daemon behavior.
From what I can recall, it is less racy for the process doing
bind(2) to unlink it if stale.
Eric Wong [Thu, 14 Jan 2021 07:06:16 +0000 (19:06 -1200)]
daemon+watch: fix localization of %SIG for non-signalfd users
It turns out "local" did not take effect in the way we used it:
http://nntp.perl.org/group/perl.perl5.porters/258784
<CAHhgV8hPbcmkzWizp6Vijw921M5BOXixj4+zTh3nRS9vRBYk8w@mail.gmail.com>
Fortunately, none of the old use cases seem affected, unlike the
previous lei change to ensure consistent SIGPIPE handling.
Eric Wong [Thu, 14 Jan 2021 07:06:15 +0000 (19:06 -1200)]
lei: test SIGPIPE, stop xsearch workers on client abort
The new test ensures consistency between oneshot and
client/daemon users. Cancelling an in-progress result now also
stops xsearch workers to avoid wasted CPU and I/O.
Note the lei->atfork_child_wq usage changes, it is to workaround
a bug in Perl 5: http://nntp.perl.org/group/perl.perl5.porters/258784
<CAHhgV8hPbcmkzWizp6Vijw921M5BOXixj4+zTh3nRS9vRBYk8w@mail.gmail.com>
This switches the internal protocol to use SOCK_SEQPACKET
AF_UNIX sockets to prevent merging messages from the daemon to
client to run pager and kill/exit the client script.
Eric Wong [Sun, 10 Jan 2021 12:15:19 +0000 (12:15 +0000)]
lei: query: restore JSON output overview
This internal API is better suited for fork-friendliness (but
locking + dedupe still needs to be re-added).
Normal "json" is the default, though stream-friendly "concatjson"
and "jsonl" (AKA "ndjson" AKA "ldjson") all seem working
(though tests aren't working, yet).
For normal "json", the biggest downside is the necessity of a
trailing "null" element at the end of the array because of
parallel processes, since (AFAIK) regular JSON doesn't allow
trailing commas, unlike JavaScript.
Eric Wong [Sun, 10 Jan 2021 12:15:18 +0000 (12:15 +0000)]
lei_xsearch: transfer 4 FDs internally, drop IO::FDPass
It's easier to make the code more generic by transferring
all four FDs (std(in|out|err) + socket) instead of omitting
stdin.
We'll be reading from stdin on some imports, and possibly
outputting to stdout, so omitting stdin now would needlessly
complicate things.
The differences with IO::FDPass "1" code paths and the "4"
code paths used by Inline::C and Socket::MsgHdr are far too
much to support and test at the moment.
Eric Wong [Sun, 10 Jan 2021 12:15:17 +0000 (12:15 +0000)]
lei: run pager in client script
While most single keystrokes work fine when the pager is
launched from the background daemon, Ctrl-C and WINCH can cause
strangeness when connected to the wrong terminal.
Eric Wong [Sun, 10 Jan 2021 12:15:16 +0000 (12:15 +0000)]
lei: fork + FD cleanup
Do a better job of closing FDs that we don't want shared with
the work queue workers. We'll also fix naming and use
"atfork_prepare" instead of "atfork_parent" to match
pthread_atfork(3) naming.
Eric Wong [Sun, 10 Jan 2021 12:15:15 +0000 (12:15 +0000)]
lei: get rid of client {pid} field
Using kill(2) is too dangerous since extremely long
queries may mean the original PID of the aborted lei(1)
client process to be recycled by a new process. It would
be bad if the lei_xsearch worker process issued a kill
on the wrong process.
So just rely on sending the exit message via socket.
Eric Wong [Sun, 10 Jan 2021 12:15:14 +0000 (12:15 +0000)]
ipc: drop unused fields, default sighandlers for wq
Relying on signal handlers to kill a particular worker was a
laggy/racy idea and I gave up on the idea of targetting workers
explicitly and instead chose to make wq_worker_decr stop the
next idle worker ->wq_exit.
We will however attempt to support sending signals to
a process group.
Eric Wong [Sun, 10 Jan 2021 12:15:13 +0000 (12:15 +0000)]
ipc: fix IO::FDPass use with a worker limit of 1
IO::FDPass is our last choice for implementing the workqueue
because its lack of atomicity makes it impossible to guarantee
all requests of a single group hit a single worker out of many.
So the only way to use IO::FDPass for workqueues it to only have
a single worker. A single worker still buys us a small amount
of parallelism because of the parent process.
Eric Wong [Sun, 10 Jan 2021 12:15:12 +0000 (12:15 +0000)]
ipc: start supporting sending/receiving more than 3 FDs
Actually, sending 4 FDs will be useful for lei internal xsearch
work once we start accepting input from stdin. It won't be used
with the lightweight lei(1) client, however.
Eric Wong [Sun, 10 Jan 2021 12:15:11 +0000 (12:15 +0000)]
lei: query: ensure pager exit is instantaneous
Improve interactivity and user experience by allowing the user
to return to the terminal immediately when the pager is exited
(e.g. hitting the `q' key in less(1)).
This is a massive change which restructures query handling to
allow parallel search when --thread expansion is in use and
offloading to a separate worker when --thread is not in use.
The Xapian query offload changes allow us to reenter the event
loop right away once the search(es) are shipped off to the work
queue workers.
This means the main lei-daemon process can forget the lei(1)
client socket immediately once it's handed off to worker
processes.
We now unblock SIGPIPE in query workers and send an exit(141)
response to the lei(1) client socket to denote SIGPIPE.
This also allows parallelization for users using "lei q" from
multiple terminals.
JSON output is currently broken and will need to be restructured
for more flexibility and fork-safety.
Eric Wong [Sun, 10 Jan 2021 12:15:10 +0000 (12:15 +0000)]
lei: fix oneshot TTY detection by passing STD*{GLOB}
... instead of STD*{IO}. I'm not sure why *STDOUT{IO} being an
IO::File object disqualifies it from the "-t" perlop check
returning true on TTY, but it does. So use *STDOUT{GLOB} for
now.
Eric Wong [Sun, 10 Jan 2021 12:15:04 +0000 (12:15 +0000)]
ipc: work queue support via SOCK_SEQPACKET
This will allow any number of younger sibling processes to
communicate with older siblings directly without relying on a
mediator process. This is intended to be useful for
distributing search work across multiple workers without caring
which worker hits it (we only care about shard members).
And any request sent with this will be able to hit any worker
without locking on our part.
Unix stream sockets with a listener were also considered;
binding to a file on the FS may confuse users given there's
already a socket path for lei(1). Linux-only Abstract or
autobind sockets are rejected due to lack of portability.
SOCK_SEQPACKET via socketpair(2) was chosen since it's POSIX
2008 and available on FreeBSD 9+ in addition to Linux, and
doesn't require filesystem access.
Eric Wong [Sun, 10 Jan 2021 12:15:03 +0000 (12:15 +0000)]
ipc: avoid excessive evals
We should not need an eval for warning with our code base.
Nowadays, dwaitpid() automatically does the right thing
regardless of whether we're in the event loop, so no eval
is needed there, either.
Eric Wong [Sun, 10 Jan 2021 12:15:02 +0000 (12:15 +0000)]
cmd_ipc: send FDs with buffer payload
For another step in in syscall reduction, we'll support
transferring 3 FDs and a buffer with a single sendmsg/recvmsg
syscall using Socket::MsgHdr if available.
Beyond script/lei itself, this will be used for internal IPC
between search backends (perhaps with SOCK_SEQPACKET). There's
a chance this could make it to the public-facing daemons, too.
This adds an optional dependency on the Socket::MsgHdr package,
available as libsocket-msghdr-perl on Debian-based distros
(but not CentOS 7.x and FreeBSD 11.x, at least).
Our Inline::C version in PublicInbox::Spawn remains the last
choice for script/lei due to the high startup time, and
IO::FDPass remains supported for non-Debian distros.
Since the socket name prefix changes from 3 to 4, we'll also
take this opportunity to make the argv+env buffer transfer less
error-prone by relying on argc instead of designated delimiters.
Eric Wong [Sun, 10 Jan 2021 12:15:01 +0000 (12:15 +0000)]
ipc: add support for asynchronous callbacks
Similar to git->cat_async, this will let us deal with responses
asynchronously, as well as being able to mix synchronous and
asynchronous code transparently (though perhaps not optimally).
Eric Wong [Sun, 10 Jan 2021 12:15:00 +0000 (12:15 +0000)]
ds: block signals when reaping
This lets us call dwaitpid long before a process exits
and not have to wait around for it.
This is advantageous for lei where we can run dwaitpid on the
pager as soon as we spawn it, instead of waiting for a client
socket to go away on DESTROY.
Eric Wong [Sun, 10 Jan 2021 12:14:58 +0000 (12:14 +0000)]
lei query + pagination sorta working
Parallelism and interactivity with pager + SIGPIPE needs work;
but results are shown and phrase search works without shell
users having to apply Xapian quoting rules on top of standard
shell quoting.
Eric Wong [Tue, 5 Jan 2021 01:29:10 +0000 (01:29 +0000)]
v2writable: exact discontiguous history handling
We've always temporarily unindexeded messages before reindexing
them again if there's discontiguous history.
This change improves the mechanism we use to prevent NNTP and
IMAP clients from seeing duplicate messages.
Previously, we relied on mapping Message-IDs to NNTP article
numbers to ensure clients would not see the same message twice.
This worked for most messages, but not for for messages with
reused or duplicate Message-IDs.
Instead of relying on Message-IDs as a key, we now rely on the
git blob object ID for exact content matching. This allows
truly different messages to show up for NNTP|IMAP clients, while
still those clients from seeing the message again.
Eric Wong [Tue, 5 Jan 2021 09:04:37 +0000 (09:04 +0000)]
address: pairs: new helper for JMAP (and maybe lei)
Per JMAP RFC 8621 sec 4.1.2.3, we should be able to
denote the lack of a phrase/comment corresponding to an
email address with a JSON "null" (or Perl `undef').
Eric Wong [Tue, 5 Jan 2021 09:04:36 +0000 (09:04 +0000)]
lei: use client env as-is, drop daemon-env command
There may be subtle misbehaviours when mixing the existing
daemon env and the client-supplied env. Just do the simplest
thing and use the client env as-is.
We'll also start the ->event_step callback since we'll need
to remember some things for long-lived commands.
Eric Wong [Tue, 5 Jan 2021 09:04:34 +0000 (09:04 +0000)]
lei: completion: fix filename completion
"-o default" is what we want from "complete", "-o filename" just
tells readline the result from the "_lei" function might be a
filename and quote appropriately.
Eric Wong [Mon, 4 Jan 2021 04:16:23 +0000 (04:16 +0000)]
lei: fix opt_dash to pass non-dash args to @argv
The special "<>" handling in Getopt::Long actually invokes the
callback for every single command-line arg, not just those
prefixed by "-". This will let us pass arbitrary non-dashed
words for search queries so users can type queries naturally
without quoting (unless they want phrase search).
Eric Wong [Sun, 3 Jan 2021 20:58:29 +0000 (20:58 +0000)]
lei: prefer IO::FDPass over our Inline::C recv_3fds
While our recv_3fds() implementation is more efficient
syscall-wise, loading Inline takes nearly 50ms on my machine
even after Inline::C memoizes the build. The current ~20ms in
the fast path is barely acceptable to me, and 50ms would be
unusable.
Eventually, script/lei may invoke tcc(1) or cc(1) directly in
the fast path, but it needs @INC for the slow path, at least.
We'll encode the number of FDs into the socket name allow
parallel installations, for now.
Eric Wong [Sun, 3 Jan 2021 02:06:16 +0000 (02:06 +0000)]
ipc: switch to one-way pipes
This fixes a performance regression in multi-process v2 indexing
due to the switch to PublicInbox::IPC. While Unix sockets are
fewer FDs to manage, pipes allow unprivileged processes to use
larger buffers (up to 1M) on out-of-the-box Linux instances.
A larger buffer via F_SETPIPE_SZ afforded by pipes was proven
valuable during v2 development in 2018 and continues to be
valuable when we get significant amounts of one-way traffic from
the producer parent to worker children.
Compression may be an option for systems without F_SETPIPE_SZ;
but it increases CPU usage with no memory bandwidth savings on
hosts where larger buffers are available.
Eric Wong [Sun, 3 Jan 2021 02:06:15 +0000 (02:06 +0000)]
use Eml (or MIME) objects for all indexing paths
We don't need to be keeping the raw message around after it hits
git. Shard work now relies on Storable (or Sereal) and all of
the indexing code relies on the Email::MIME-like API of Eml to
access interesting parts of the message.
Similarly, smsg->{raw_bytes} is no longer carried around and we
do the CRLF adjustment when setting smsg->{bytes}.
There's also a small simplification to t/import.t while
we're in the area to use xqx instead of spawn/popen_rd.
Eric Wong [Sun, 3 Jan 2021 02:06:12 +0000 (02:06 +0000)]
searchidxshard: use PublicInbox::IPC to kill lots of code
It's nice to prove the new code works by swapping it into
the current V2Writable / SearchIdxShard packages. This is
only the first step for the core bits, and we'll be able
to delete more code in a subsequent patch.
Eric Wong [Sun, 3 Jan 2021 02:12:06 +0000 (12:12 -1400)]
gcf2client: split out request API from regular git
While Gcf2Client is designed to mimic what git-cat-file writes
to stdout, its request format is different to support requests
with a git repository path included.
We'll highlight the distinction and make the GitAsyncCat support
code easier-to-follow as a result.
Since Gcf2Client relies on DS, we can rely on DS-specific code
here, too, and use a single Unix socket instead of separate
input and output pipes, reducing memory overhead in both users
and kernel space. Due to the interactive nature of requests and
responses, the buffer size limitations of Unix sockets on Linux
seems inconsequential here (just like it is for existing "git
cat-file --batch" use).
Eric Wong [Sun, 3 Jan 2021 11:24:51 +0000 (11:24 +0000)]
lei: fix output race in client/daemon mode
The daemon needs to flush stdout before disconnecting or killing
clients, otherwise they may reread empty data on redirected
outputs. We also don't want to unbuffer stdout too early in
case we have lots of small chunks of data to output.
The received ($self->{2}) will always have autoflush, matching normal
STDERR behavior.
Eric Wong [Sun, 3 Jan 2021 11:24:50 +0000 (11:24 +0000)]
send and receive all 3 FDs at once
We'll always be transferring stdin, stdout, and stderr together
for lei. Perhaps I lack imagination or foresight, but I can't
think of a reason to send more or less FDs.
Eric Wong [Sat, 2 Jan 2021 08:32:04 +0000 (08:32 +0000)]
lei_store: alternative unconfigured "git var" workaround
While the changes to git->qx/git->popen from commit 171a9c24022ad7ef
will be useful for the lei daemon, hiding git error messages from
actual users is probably wrong and we'll just localize GIT_*
vars for testing.
Eric Wong [Fri, 1 Jan 2021 05:47:49 +0000 (17:47 -1200)]
import: unset GIT_CONFIG with `git config --global'
GIT_CONFIG is set by -convert, and user may have it set
for other reasons. In either case, it conflicts with
any any attempt to use `git config --global` so we have
to unset it.
We need to use an absolute path after chdir in run modes
where scripts aren't loaded into in-memory subs.
The oneshot test was also failing under TEST_RUN_MODE=0 due to
no "lei-oneshot" command existing on the FS. So we force a
socket failure by making XDG_RUNTIME_DIR too large to fit into
the 108-byte .sun_path field of "struct sockaddr_un". This
even lets us simplify lei-oneshot significantly.
Eric Wong [Thu, 31 Dec 2020 13:51:54 +0000 (13:51 +0000)]
on_destroy: support PID owner guard
Since we'll be forking for Xapian indexing and maybe
other places, having a simple guard in place to ensure
OnDestroy doesn't unexpectedly unlink files or similar
is a safer option.