Eric Wong [Wed, 3 Feb 2021 21:51:44 +0000 (15:51 -0600)]
tests: guard against missing DBD::SQLite
The features we use for SharedKV could probably be implemented
with GDBM_File or SDBM_File, but that doesn't seem worth it at
the moment since we depend on SQLite elsewhere.
The new IPC stuff doesn't work without Storable or Sereal.
Storable is part of the standard library since Perl 5.8, so
we'll put a hard dependency on it for distros that package
it separately.
Data::Dumper is also part of the standard library, and
PublicInbox::MboxReader uses it, and it's frequently useful
during development.
We'll also trim down INSTALL for standard library modules so
it's hopefully less daunting for new users.
Development dependencies are noted in HACKING, now.
Email::MIME is only used for maintainer tests, so it's only
documented in HACKING.
Eric Wong [Wed, 3 Feb 2021 21:51:41 +0000 (15:51 -0600)]
HACKING: use "just-ahead-of-time" to describe Inline::C
Inline::C works during module load time, so "just-ahead-of-time"
is a better description of it than "just-in-time". I don't
think "JAOT" is a well-known enough acronym, so it's worth
spelling it out.
Eric Wong [Wed, 3 Feb 2021 08:11:42 +0000 (22:11 -1000)]
lei: use sleep(1) loop for infinite sleep
Perl may internally race and miss signals due to a lack of
self-pipe / eventfd / signalfd / EVFILT_SIGNAL usage. While our
event loop paths avoid these problems by using signalfd or
EVFILT_SIGNAL, thse sleep() calls are not within the event loop.
Eric Wong [Wed, 3 Feb 2021 08:11:41 +0000 (22:11 -1000)]
lei add-external: completion for existing URL basenames
Given the presence of one external on a certain host or prefix
path, it's logical other inboxes would share a common prefix.
For bash users, attempt to complete that using the "-o nospace"
option of bash
Eric Wong [Wed, 3 Feb 2021 08:11:38 +0000 (22:11 -1000)]
lei q: -I/--exclude/--only support globs and basenames
We can do basename matching when it's unambiguous. Since '*?[]'
characters are rare in URLs and pathnames, we'll do glob
matching by default to support a (curl-inspired) --globoff/-g
option to disable globbing.
IO::Uncompress::Gunzip seems to be losing $? when closing
PublicInbox::ProcessPipe. To workaround this, do a synchronous
waitpid ourselves to force proper $? reporting update tests to
use the new --only feature for testing invalid URLs.
This improves internal code consistency by having {pkt_op}
parse the same ASCII-only protocol script/lei understands.
We no longer pass {sock} to worker processes at all,
further reducing FD pressure on per-user limits.
Eric Wong [Tue, 2 Feb 2021 11:47:02 +0000 (11:47 +0000)]
lei q: support --jobs [SEARCHERS],[WRITERS]
This comma-delimited parameter allows controlling the number or
lei_xsearch and lei2mail worker processes. With the change
to make IPC wq_* work use the event loop, it's now safe to
run fewer worker processes for searching with no risk of
deadlocks.
MAX_PER_HOST isn't configurable yet for remote hosts,
and maybe it shouldn't be due to potential for abuse.
Eric Wong [Tue, 2 Feb 2021 11:47:01 +0000 (11:47 +0000)]
lei q: tidy up progress reporting
We won't be reporting progress when output is going to stdout
since it can clutter up the terminal unless stderr != stdout,
which probably isn't worth checking.
We'll also use a more agnostic mset_progress which may
make it easier to support worker-less invocations.
Eric Wong [Tue, 2 Feb 2021 11:46:58 +0000 (11:46 +0000)]
lei_xsearch: ensure curl.err and tail(1) cleanup happens
We can safely rely on exit(0) here when interacting with curl(1)
and git(1), unlike query workers which hit Xapian directly,
where some badness happens when hit with a signal while
retrieving an mset.
Eric Wong [Tue, 2 Feb 2021 11:46:55 +0000 (11:46 +0000)]
lei q: do not leave temporary files after oneshot exit
Avoid on-stack shortcuts which may prevent destructors from
firing since we're not inside the event loop. We'll also tidy
up the unlink mechanism in LeiOverview while we're at it.
Eric Wong [Tue, 2 Feb 2021 11:46:50 +0000 (11:46 +0000)]
lei q: support --only, --include and --exclude
-I is short for --include since it's standard for C compilers
(along with Perl and Ruby). There are no single-character
shortcuts for --exclude or --only, since I don't expect
--exclude to be used very often and --only is already short (and
will support shell completion).
Eric Wong [Tue, 2 Feb 2021 11:46:49 +0000 (11:46 +0000)]
lei q: emit progress and counting via PktOp
Sometimes it can be confusing for "lei q" to finish writing to a
Maildir|mbox and not know if it did anything. So show some
per-external progress and stats.
These can be disabled via the new --quiet/-q switch.
We differ slightly from mairix(1) here, as we use stderr
instead of stdout for reporting totals (and we support
parallel queries from various sources).
Eric Wong [Mon, 1 Feb 2021 08:28:33 +0000 (22:28 -1000)]
doc: note optional BSD::Resource use
We've actually been capable of using this since 2019(*) in our
spawn code for PSGI limiters. And it's been used since 2016 in
our tests. It's a dependency of SpamAssassin, and Danga::Socket
used it, too.
Eric Wong [Mon, 1 Feb 2021 08:28:32 +0000 (22:28 -1000)]
lei: avoid ETOOMANYREFS, cleanup imports
As with PublicInbox::IPC, we'll attempt to bump RLIMIT_NOFILE
and transparently workaround ETOOMANYREFS. If that fails,
we'll give the user a hint to bump RLIMIT_NOFILE since
ETOOMANYREFS is an uncommon error which users may be unfamiliar
with.
Eric Wong [Mon, 1 Feb 2021 08:28:31 +0000 (22:28 -1000)]
ds: next_tick: avoid $_ in top-level loop iterator
$_ at the top of a potentially deep stack below may cause
surprising behavior as I experienced with ExtSearchIdx. In the
future, we'll limit our $_ usage to easily-auditable bits (e.g.
map, grep, and small for loops)
Eric Wong [Mon, 1 Feb 2021 08:28:30 +0000 (22:28 -1000)]
ds: guard against stack-not-refcounted quirk of Perl 5
The Perl 5 stack is weakly-referenced for performance reasons.
This means it's possible for items in the stack to be freed
while executing further down the stack.
In lei (and perhaps public-facing read-only daemons in the
future), we'll fork and call PublicInbox::DS->Reset in the child
process. This causes %DescriptorMap to be clobbered, allowing
the $DescriptorMap{$fd} arg to be freed inside the child
process.
When Carp::confess or Carp::longmess is called to generate a
backtrace, it may access the @DB::args array. This array access
is not protected by reference counting and is known to cause
segfaults and other weird errors.
While the caller of an unnecessary Carp::confess may be
eliminated in a future commit, we can't guarantee our
dependencies will be free of @DB::args access attempts
in the future.
So guard against this Perl 5 quirmk by defensively bumping the
refcount of any object we call ->event_step on.
Eric Wong [Mon, 1 Feb 2021 08:28:27 +0000 (22:28 -1000)]
lei_to_mail: reduce spew on Maildir removal
At most, we'll only warn once per worker when a Maildir
disappears from under us. We'll also use the '!' OpPipe
to note the exceptional condition, and use '|' to SIGPIPE
so it'll be a bit easier for hackers to remember.
Eric Wong [Mon, 1 Feb 2021 08:28:25 +0000 (22:28 -1000)]
lei: increase initial timeout
PublicInbox::Listener unconditionally sets O_NONBLOCK upon
accept(), so we need a larger timeout under heavy load since
there's no "dataready" accept filter on the listener.
With O_NONBLOCK already set, we don't have to set it at
->event_step_init
Eric Wong [Mon, 1 Feb 2021 08:28:24 +0000 (22:28 -1000)]
sharedkv: lock and explicitly disconnect {dbh}
It may be possible for updates or changes to be uncommitted
until disconnect, so we'll use flock() as we do elsewhere
to avoid the polling retry behavior of SQLite.
We also need to clear CachedKids before disconnecting to
to avoid warnings like:
->disconnect invalidates 1 active statement handle
(either destroy statement handles or call finish on
them before disconnecting)
Eric Wong [Mon, 1 Feb 2021 08:28:20 +0000 (22:28 -1000)]
lei: keep $lei around until workers are reaped
This prevents SharedKV->DESTROY in lei-daemon from triggering
before DB handles are closed in lei2mail processes. The
{each_smsg_not_done} pipe was not sufficient in this case:
that gets closed at the end of the last git_to_mail callback
invocation.
Eric Wong [Mon, 1 Feb 2021 08:28:17 +0000 (22:28 -1000)]
ipc: more helpful ETOOMANYREFS error messages
ETOOMANYREFS is probably a unfamiliar error to most users, so
give a hint about RLIMIT_NOFILE. This can be hit on my system
running 3 simultaneous queries with my system default limit of
1024.
There's also no need to import Errno constants for uncommon
errors, so we'll stop using Errno, here.
We'll also try to bump RLIMIT_NOFILE as much as possible
to avoid this error.
Eric Wong [Mon, 1 Feb 2021 08:28:16 +0000 (22:28 -1000)]
lei: remove SIGPIPE handler
It doesn't save us any code, and the action-at-a-distance
element was making it confusing to track down actual problems.
Another potential problem was keeping references alive too long.
So do like we would a C100K server and check every write
while still ensuring lei(1) exit with a proper SIGPIPE
iff needed.
Eric Wong [Mon, 1 Feb 2021 08:28:14 +0000 (22:28 -1000)]
ipc: switch wq to use the event loop
This will let us to maximize the capability of our asynchronous
git API. This lets us avoid relying on EOF to notify lei2mail
workers; thus giving us the option of running fewer lei_xsearch
worker processes in parallel than local sources.
I tried using a synchronous git API; and even with libgit2 in
the same process to avoid the IPC cost failed to match the
throughput afforded by this change. This is because libgit2 is
built (at least on Debian) with the SHA-1 collision code enabled
and ubc_check stuff was dominating my profiles.
Kyle Meyer [Mon, 1 Feb 2021 05:57:02 +0000 (00:57 -0500)]
doc: start manpages for lei commands
Add manpages for lei and the currently implemented subcommands. The
included options and their descriptions follow to a large degree the
--help output, dropping some options that are not currently wired up.
Eric Wong [Sat, 30 Jan 2021 05:41:09 +0000 (23:41 -0600)]
content_hash: skip Sender for cross posted messages
This regression was introduced long ago and matches behavior
originally specified in the comments. It makes a noticeable
improvement with search results using -extindex ("all") and
lei results with multiple inboxes.
Update some style bits at the top of the test case while
we're at it.
Eric Wong [Fri, 29 Jan 2021 07:42:57 +0000 (12:42 +0500)]
lei: less error-prone FD mapping
Keeping track of non-standard FDs gets tricky, so make it easier
by relying on st_dev/st_ino mapping in the transmitted objects.
We'll keep using numbers for the standard FDs since we need to
be able to easily redirect them in the producer (main daemon)
process for (gzip|bzip2|xz) if writing to a compressed mbox.
Eric Wong [Fri, 29 Jan 2021 07:42:56 +0000 (12:42 +0500)]
ipc: more consistent behavior between worker types
Localize signals inside the respective worker loops
in case there's circular references.
We'll also rely on OnDestroy to trigger exits from the
ipc_worker_loop like we do with wq_worker_loop. And
also add some more developer documentation to help future
developers.
The default signals remain different, for now.
Cleanup some unnecessary "use" statements while we're
loading OnDestroy.
Eric Wong [Wed, 27 Jan 2021 09:42:25 +0000 (03:42 -0600)]
lei: set PWD correctly for path expansion
While commit d1b9582872d1824f166a038dcf32b6ae8c6dc735
("lei: pass FD to CWD via cmsg, use fchdir on server")
ensured things work properly to get the daemon in the
right directory, it forgot to deal with places where
we expand relative paths based on the current working
directory.
Eric Wong [Mon, 25 Jan 2021 06:41:57 +0000 (22:41 -0800)]
spawn: split() on regexp, not a literal string
It doesn't appear Perl (as of 5.32.x) has any internal
optimization for splitting on a single-byte, so give it
a regexp instead of letting it compile and discard a
new one every single time.
Eric Wong [Mon, 25 Jan 2021 06:41:56 +0000 (22:41 -0800)]
miscidx: switch to lazy transactions
This fixes a sporadic failure on a 1/2 core VM where
"git cat-file --batch" hasn't started up by the time
$cleanup->() destroys the ALL.git directory in t/lei.t
(but not t/lei-oneshot.t).
This happens because dwaitpid() runs inside the event loop
asynchronously and we were able to return to the client before
the cat-file process could even start.
I could not reproduce this failure on my usual 4-core
workstation via "schedtool -a 0x1" to force the entire
test to use a single core.
Lazy transactions matches OverIdx and SearchIdx behavior, and
I've verified this lets us avoid problems with old Xapian
versions (on CentOS 7.x) which failed to set FD_CLOEXEC.
Eric Wong [Mon, 25 Jan 2021 04:53:46 +0000 (19:53 -0900)]
doc: start working on public-inbox-extindex(1) manpage
It's barely started, but I started writing this weeks ago, but
I'm still unsure about some behavioral/usability things and
hoping work on lei(1) can flush them out.
Eric Wong [Mon, 25 Jan 2021 01:18:57 +0000 (17:18 -0800)]
lei q: continue remote search if torsocks(1) is missing
torsocks is just one of many ways to get curl to use Tor,
so we'll continue if we can't find torsocks in our PATH
and assume the user has a proxy configured via curlrc,
the command-line, environment variable, or even firewall
rules.
Eric Wong [Mon, 25 Jan 2021 01:18:55 +0000 (17:18 -0800)]
lei q: demangle and quiet curl output
curl(1) writes to stderr one byte-at-a-time (presumably for the
progress bar). This ends up being unreadable on my terminal
when parallel processes are trying to write error messages.
So instead, we'll capture the output to a file and run
'tail -f' on it if --verbose is enabled.
Since HTTP 404s from non-existent results are a common response,
we'll ignore them and stay silent, matching behavior of local
searches.
Eric Wong [Mon, 25 Jan 2021 01:18:54 +0000 (17:18 -0800)]
lei q: drop "oid" output format
The default deduplication command-line arguments would be
non-sensical for such an option and probably confusing. It
doesn't seem worth the code to support OID-only output when it's
easy enough to use one of the JSON formats to extract the same
info.
We also don't have OIDs if using remotes, and the
to-be-implemented memoization will be optional.
Eric Wong [Mon, 25 Jan 2021 01:18:53 +0000 (17:18 -0800)]
lei: reinstate JSON smsg output deduplication
This was accidentally clobbered completely in
("lei q: fix JSON overview with remote externals").
There are now more tests to prevent future regressions.
Eric Wong [Sun, 24 Jan 2021 11:46:55 +0000 (04:46 -0700)]
smsg: parse_references: micro-optimization
With Perl 5.10+, we can rely on the defined-or-assignment (//=)
operator to avoid repeatedly rewriting an SV.
This may not provide a measurable difference here, but
it's more consistent with current style where things like
commit a05445fb400108e60ede7d377cf3b26a0392eb24
("config: config_fh_parse: micro-optimize") provide a measurable
improvement.
Eric Wong [Sun, 24 Jan 2021 11:46:53 +0000 (04:46 -0700)]
lei q: fix JSON overview with remote externals
We can't (and don't need to) repeatedly get the $each_smsg
callback for each URI since that clobbers {ovv_buf} before
it can be output.
I initially thought this was a dedupe-related bug and
moved the dedupe code into the $each_smsg callback to
minimize differences. Nevertheless it's a nice code
reduction.
I also thought it was related to incomplete smsg info,
so {references} is now filled in correctly for dedupe.
Eric Wong [Sun, 24 Jan 2021 11:46:51 +0000 (04:46 -0700)]
lei q: honor --no-local to force remote searches
This can be useful for testing remote behavior, or for
augmenting local results. It'll also be possible to explicitly
include/exclude externals via CLI switches (once names are
decided).
Eric Wong [Sun, 24 Jan 2021 11:46:49 +0000 (04:46 -0700)]
ipc: get rid of wq_set_recv_modes
Just open every FD as read/write. Perl (or any non-broken
runtime) won't care and won't attempt to use F_SETFL to alter
file description flags; as attempting to change those would
lead to unpleasant side effects if the file description is
shared with another process.
Eric Wong [Sun, 24 Jan 2021 11:46:48 +0000 (04:46 -0700)]
ipc: wq supports arbitrarily large payloads
This should not be needed, but somebody using lei could
theoretically create thousands of external URLs and
only have a handful of workers, which means the per-worker
URI list could be large.