My installation recently somehow ended up with a recording with a
duration of 503793844 90,000ths of a second, way over the maximum of 5
minutes. (Looks like the machine was pretty unresponsive at the time
and/or having network problems.)
When this happens, the system really spirals. Every flush afterward (12
per minute with my installation) fails with a CHECK constraint failure
on the recording table. It never gives up on that recording. /var/log
fills pretty quickly as this failure is extremely verbose (a stack
trace, and a line for each byte of video_index). Eventually the sample
file dirs fill up too as it continues writing video samples while GC is
stuck. The video samples are useless anyway; given that they're not
referenced in the database, they'll be deleted on next startup.
This ensures the offending recording is never added to the database, so
we don't get the same persistent problem. Instead, writing to the
recording will fail. The stream will drop and be retried. If the
underlying condition that caused a too-long recording (many
non-key-frames, or the camera returning a crazy duration, or the
monotonic clock jumping forward extremely, or something) has gone away,
the system should recover.
This is so far completely untested, for use by a new UI prototype.
It creates a new URL endpoint which sends one video/mp4 media segment
per key frame, with the dependent frames included. This means there will
be about one key frame interval of latency (typically about a second).
This seems hard to avoid, as mentioned in issue #59.
I went with the third idea in 1ce52e3: have the tests run each iteration
of the syncer explicitly. These are messy tests that know tons of
internal details, but I think they're less confusing and racy than if I
had the syncer running in a separate thread.
Now each syncer has a binary heap of the times it plans to do a flush.
When one of those times arrives, it rechecks if there's something to do.
Seems more straightforward than rechecking each stream's first
uncommitted recording, especially with the logic to retry failed flushes
every minute.
Also improved the info! log for each flush to see the actual recordings
being flushed for better debuggability.
No new tests right now. :-( They're tricky to write. One problem is that
it's hard to get the timing right: a different flush has to happen
after Syncer::save's database operations and before Syncer::run calls
SimulatedClocks::recv_timeout with an empty channel[*], advancing the
time. I've thought of a few ways of doing this:
* adding a new SyncerCommand to run something, but it's messy (have
to add it from the mock of one of the actions done by the save),
and Box<dyn FnOnce() + 'static> not working (see
rust-lang/rust#28796) makes it especially annoying.
* replacing SimulatedClocks with something more like MockClocks.
Lots of boilerplate. Maybe I need to find a good general-purpose
Rust mock library. (mockers sounds good but I want something that
works on stable Rust.)
* bypassing the Syncer::run loop, instead manually running iterations
from the test.
Maybe the last way is the best for now. I'm likely to try it soon.
[*] actually, it's calling Receiver::recv_timeout directly;
Clocks::recv_timeout is dead code now? oops.
* remove intermediate bool from adjust_day.
* rewrite LockedDatabase::list_aggregate_recordings.
I started by collapsing the flush into the first part of the if, in a
similar way to adjust_day. But then I refactored more and ended up
with a structure that probably would have been allowed with the old
lexical borrow checker. I think it's more readable, and it does 1
btree operation per row where before it did 2 or 3.
This is mostly just "cargo fix --edition" + Cargo.toml changes.
There's one fix for upgrading to NLL in db/writer.rs:
Writer::previously_opened wouldn't build with NLL because of a
double-borrow the previous borrow checker somehow didn't catch.
Restructure to avoid it.
I'll put elective NLL changes in a following commit.
I initially chose SameSite=Lax because I thought if a user followed a
link to the landing page, the landing page's ajax requests wouldn't send
the cookie. But I just did an experiment, and that's not true. Only the
initial page load (of a .html file) lacks the cookie. All of its
resources and ajax requests send the cookie. I'm not sure about
document.cookie accesses, but my cookie is HttpOnly anyway, so it's
irrelevant. So no reason to be lax.
I just ran a "cargo test" on this after a round of tweaks, not
"cargo test --all", so I missed compile errors in the db crate,
and a Javascript lint config error. travis-ci caught these.
Some caveats:
* it doesn't record the peer IP yet, which makes it harder to verify
sessions are valid. This is a little annoying to do in hyper now
(see hyperium/hyper#1410). The direct peer might not be what we want
right now anyway because there's no TLS support yet (see #27). In
the meantime, the sane way to expose Moonfire NVR to the Internet is
via a proxy server, and recording the proxy's IP is not useful.
Maybe better to interpret a RFC 7239 Forwarded header (and/or
the older X-Forwarded-{For,Proto} headers).
* it doesn't ever use Secure (https-only) cookies, for a similar reason.
It's not safe to use even with a tls proxy until this is fixed.
* there's no "moonfire-nvr config" support for inspecting/invalidating
sessions yet.
* in debug builds, logging in is crazy slow. See libpasta/libpasta#9.
Some notes:
* I removed the Javascript "no-use-before-defined" lint, as some of
the functions form a cycle.
* Fixed#20 along the way. I needed to add support for properly
returning non-OK HTTP statuses to signal unauthorized and such.
* I removed the Access-Control-Allow-Origin header support, which was
at odds with the "SameSite=lax" in the cookie header. The "yarn
start" method for running a local proxy server accomplishes the same
thing as the Access-Control-Allow-Origin support in a more secure
manner.
This is a minor code size reduction - instead of being monomorphized
into four variants (according to "cargo llvm-lines"), it's now
monomorphized into two. The stripped release binary on macOS is about
8kB smaller (0.15%). Not a huge improvement but better than nothing.
Benchmarks seem unchanged (though they have a lot of variance).
There was a race condition here because it wasn't waiting for the db
flush to complete. This made write_path_retries sometimes not reflect
the consequence of the flush, causing an assertion failure. I assume it
was also responsible for gc_path_retries timeouts under travis-ci.
The new behavior eliminates a couple unpleasant edge cases in which it
would never flush:
* if all recording stops, whatever was unflushed would stay that way
* if every recording attempt produces a 0-duration recording (such as if the
camera sends only one frame and thus no PTS delta can be calculated),
the list of recordings to flush would continue to grow
I moved the clocks member from LockedDatabase to Database to make this happen,
so the new DatabaseGuard (replacing a direct MutexGuard<LockedDatabase>) can
access it before acquiring the lock.
I also made the type of clock a type parameter of Database (and so several
other things throughout the system). This allowed me to drop the Arc<>, but
more importantly it means that the Clocks trait doesn't need to stay
object-safe. I plan to take advantage of that shortly.
--features=bundled enables -DSQLITE_DEFAULT_FOREIGN_KEYS=1, and so some
operations have to be done in the proper order.
* enable foreign key enforcement all the time, so I test this more reliably.
* reorder some parts of the v1->v3 order. foreign key enforcement is
immediate (rather than deferred) by default. and ensure
old_recording_playback isn't left with a dangling reference to old_recording
at the v2 stage. Instead, wait until v3 to delete tables it depends on.
This is only the database schema, which I'm adding now in the hopes of
freezing schema version 3. There's no way yet to create users, much less
actually authenticate.
These are not actually populated by the code yet. I'm trying to get the
v3 schema frozen as soon as possible; actually using the fields can come
later.
Add some explanation of their value in time.md, along with some general
musing on leap seconds, and a correction on the frequency error of my cameras.
* separate these out into a new file, writer.rs, as dir.rs was getting
unwieldy.
* extract traits for the parts of SampleFileDir and std::fs::File they needed;
set up mock implementations.
* move clock.rs to a new base crate to be accessible from the db crate.
* add tests that exercise all the retry paths.
* bugfix: account for the new recording's bytes when calculating how much to
delete.
* bugfix: when retrying an unlink failure in collect_garbage, we shouldn't
warn about all the recordings no longer existing. Do this by retrying each
step rather than the whole procedure again.
* avoid double-panic scenarios, which I hit while tweaking the mocks. These
are quite annoying to debug as Rust doesn't print information about either
panic. I ended up using lldb to get a backtrace. Better to be cautious about
what we're doing when already panicking.
* give more context on raw::insert_recording errors, which I hit as well while
tweaking the new tests.
There may be considerable lag between being fully written and being committed
when using the flush_if_sec feature. Additionally, this is a step toward
listing and viewing recordings before they're fully written. That's a
considerable delay: 60 to 120 seconds for the first recording of a run,
0 to 60 seconds for subsequent recordings.
These recordings aren't yet included in the information returned by
/api/?days=true. They probably should be, but small steps.
I want to start having the db.rs version augment this with the uncommitted
recordings, and it's nice to have the separation of the raw db vs augmented
versions. Also, this fits with the general theme of shrinking db.rs a bit.
I had to put the raw video_sample_entry_id into the rows rather than
the video_sample_entry Arc. In hindsight, this is better anyway: the common
callers don't need to do the btree lookup and arc clone on every row. I think
I'd originally done it that way only because I was quite new to rust and
didn't understand that db could be used from within the row callback given
that both borrows are immutable.
This was considering them as 0, so it would under-delete until the next flush
them delete all at once. That effectively doubled the number of bytes not yet
deleted as they're first transferred to garbage, flushed again, then unlinked.
In hindsight, the "post_tx" step in the upgrade process introduced in
e7f5733 doesn't make sense. If the procedure fails at this stage, nothing says
it still needs to be completed. If the sample file dirs have to be updated
after the database, then there should be another database version to mark that
it's fully completed, and indeed that's the purpose version 3 serves. So get
rid of the Upgrader trait and just go back to a simple run function per
version.
In the case of the sample file dir metadata, it actually can happen before the
database transaction; the stuff written to the database later just needs to be
consistent with what it finds if there's an existing metadata file from a
half-completed update.
For safety, ensure there are no unexpected directory contents before
upgrading 1->2, and ensure the metadata matches before upgrading 2->3.
I want to be able to use it in etags without having to do a full scan of the
recording_playback in advance, which would greatly increase time to first
byte. I probably will even use it in urls to ensure the segments they point to
are stable. I haven't actually done this yet - it will wait until I implement
serving unflushed recordings - but I want to get the schema set up properly.
Every recording it starts must be sent to the syncer with at least one sample
written. It will try forever (unless the channel is down, then panic). This
avoids the situation in which it prevents something in the uncommitted
VecDeque from ever being synced and thus any further recordings from being
flushed.
The new approach is to, rather than panicking, retry forever. The assumption
is that if a given operation is failing, a following operation is unlikely to
succeed, so it's simpler to just keep trying the earlier one than come up with
ways to undo it and proceed with later operations.
I still need to apply this approach to the Writer class. It currently unwraps
(crashes) or just gives up on a recording without ever sending it to the
Syncer. Given that recordings are all synced in order, that means further ones
can never be synced.
When list_oldest_recordings was called twice with no intervening flush, it
returned the same rows twice. This led to trying to delete it twice and all
following flushes failing with a "no such recording x/y" message. Now, return
each row only once, and track how many bytes have been returned.
I think dir.rs's logic is still wrong for how many bytes to delete when
multiple recordings are flushed at once (it ignores the bytes added by the
first when computing the bytes to delete for the second), but this is
progress.
I mistakenly thought these had to be monomorphized. (The FnOnce still
does, until rust-lang/rfcs#1909 is implemented.) Turns out this way works
fine. It should result in less compile time / code size, though I didn't check
this.
This needs a separate run of "cargo +nightly bench --features=nightly", so I
missed it in a couple previous commits. I probably should set up travis-ci...
As noted in schema.sql, this can be used for disambiguation. It also may be
useful in diagnosing data integrity problems.
Also, sneak in a couple minor improvements: better diagnostics in a couple
places, fix to 1->2 upgrade procedure.
This improves the practicality of having many streams (including the doubling
of streams by having main + sub streams for each camera). With these tuned
properly, extra streams don't cause any extra write cycles in normal or error
cases. Consider the worst case in which each RTSP session immediately sends a
single frame and then fails. Moonfire retries every second, so this would
formerly cause one commit per second per stream. (flush_if_sec=0 preserves
this behavior.) Now the commits can be arbitrarily infrequent by setting
higher values of flush_if_sec.
WARNING: this isn't production-ready! I hacked up dir.rs to make tests pass
and "moonfire-nvr run" work in the best-case scenario, but it doesn't handle
errors gracefully. I've been debating what to do when writing a recording
fails. I considered "abandoning" the recording then either reusing or skipping
its id. (in the latter case, marking the file as garbage if it can't be
unlinked immediately). I think now there's no point in abandoning a recording.
If I can't write to that file, there's no reason to believe another will work
better. It's better to retry that recording forever, and perhaps put the whole
directory into an error state that stops recording until those writes go
through. I'm planning to redesign dir.rs to make this happen.
It should reduce compile time / memory usage to put quite a bit of the code
into a separate crate. I also intend to limit visibility of some things to
only within the db crate, but that's for a future change. This is the smallest
move that will compile.
