* have a timeout for opening the connection and getting the next
video frame. The former is quite important. The latter is arguably
redundant with the keepalive timer, but this ensures we actually
get a full frame in this timespan rather than some keepalive
responses, RTCP sender reports, or partial frames.
* don't drop extra stuff on loss; just note it. I'm not sure what the
right behavior is but I think I shouldn't change too much at once.
I see a lot of yields and such in CPU profiles. I think the workers
are frequently waking up, finding there's not much to do, and going back
to sleep. Reducing the number of worker threads seems reasonable.
Moonfire NVR has some enforcement on its own; this makes retina vs
ffmpeg more of an apples-to-apples comparison.
I'm also thinking of dropping enforcement from retina; enough things
have sketchy timestamps that this policy doesn't make much sense anyway.
This isn't well-tested and doesn't yet support an initial connection
timeout. But in a quick test, it successfully returns video!
I'd like to do some more aggressive code restructuring for zero-copy
and to have only one writer thread per sample file directory (rather
than the syncer thread + one writer thread per RTSP stream). But I'll
likely wait until I drop support for ffmpeg entirely.
This is (slightly) complicating the switch from ffmpeg to retina
as the RTSP client. And it's not really that close to what I want
to end up with for analytics:
* I'd prefer the analytics happen in a separate process for
several reasons
* Feeding the entire frame to the object detector doesn't produce
good results.
* It doesn't do anything with the results yet anyway.
This caused served chunks to be truncated. On seek, nginx sometimes
served 502 errors, chrome sometimes returned
ERR_CONTENT_LENGTH_MISMATCH, and videos weren't playing properly.
Reading from the mmap()ed region in the tokio threads could cause
them to stall:
* That could affect UI serving when there were concurrent
UI requests (i.e., not just requests that needed the reads in
question anyway).
* If there's a faulty disk, it could cause the UI to totally hang.
Better to not mix disks between threads.
* Soon, I want to handle RTSP from the tokio threads (#37). Similarly,
we don't want RTSP streaming to block on operations from unrelated
disks.
I went with just one thread per disk which I think is sufficient.
But it'd be possible to do a fixed-size pool instead which might improve
latency when some pages are already cached.
I also dropped the memmap dependency. I had to compute the page
alignment anyway to get mremap to work, and Moonfire NVR already is
Unix-specific, so there wasn't much value from the memmap or memmap2
crates.
Fixes#88
* my dad's GW4089IP cameras use 720x480
* some Reolink cameras use 640x352
* I'm playing with rotated cameras (16x9 -> 9x16)
I'd prefer to calculate pasp from a configured camera aspect ratio
than to hardcode the assumption these are 16x9, but that requires
a schema change. This is an improvement for now.
The immediate motivation is to address these CI failures with nightly:
https://github.com/scottlamb/moonfire-nvr/runs/2593322801?check_suite_focus=true
```
Compiling lock_api v0.4.2
error[E0557]: feature has been removed
--> /home/runner/.cargo/registry/src/github.com-1ecc6299db9ec823/lock_api-0.4.2/src/lib.rs:91:42
|
91 | #![cfg_attr(feature = "nightly", feature(const_fn))]
| ^^^^^^^^ feature has been removed
|
= note: split into finer-grained feature gates
```
Strangely, they don't occur locally with "rustc 1.54.0-nightly
(fe72845f7 2021-05-16)" but do on CI with the exact same version?!?
I don't get it, but lock-api 0.4.4 is advertised as being updated for
latest nightly, so I expect this will address the problem anyway.
I saw this error once:
Apr 27 21:01:33 nuc moonfire-nvr[188570]: s-reolink-sub
moonfire_nvr::streamer] reolink-sub: sleeping for Duration { secs: 1,
nanos: 0 } after error: CHECK constraint failed: video_sample_entry
and would like to understand it better.
* API change: in update signals, allow setting a start time relative
to now. This is an accuracy improvement in the case where the client
has been retrying an initial request for a while. Kind of an obscure
corner case but easy enough to address. And use a more convenient
enum representation.
* in update signals, choose `now` before acquiring the database lock.
If lock acquisition takes a long time, this more accurately reflects
the time the caller intended.
* in general, make Time and Duration (de)serializable and use them
in json types. This makes the types more self-describing, with
better debug printing on both the server side and on the client
library (in moonfire-playground). To make this work, base has to
import serde which initially seemed like poor layering to me, but
serde seems to be imported in some pretty foundational Rust crates
for this reason. I'll go with it.
In particular, this was happening out of the box on Raspberry Pi OS Lite
20210304, as reported by ironoxidizer@gmail.com here:
https://groups.google.com/g/moonfire-nvr-users/c/2j9LvfFl2u8/m/tJcNS2WfCQAJ
* adjust main.rs to make the problem more obvious
* mention it in the troubleshooting guide
* sidestep it in the nvr docker wrapper script
also just use --networking=host rather than --publish (avoiding a proxy
process). I'm using Docker to simplify the build and deployment process,
not as a security boundary, so just do the simpler thing.
To improve reliability of live streams (#59) on Safari.
Safari was dropping the cookie from websocket update requests.
(But it worked sometimes. I don't get why.) I saw folks on the Internet
thinking this related to HttpOnly:
* https://developer.apple.com/forums/thread/104488
* https://stackoverflow.com/q/47742807/23584
but I still see this behavior without HttpOnly. SameSite=Strict vs
SameSite=Lax appears to make a difference. Try that instead.
SameSite=Strict is pointless for us anyway as noted in a new comment.
Turning off HttpOnly would be more unfortunate security-wise.
As required for live view (#59) to work on Safari.
Safari has some "interesting" expectations:
* There must be a non-empty list of compatible brands. The major brand
is not automatically included. (Looks like ISO/IEC 14496-12 doesn't
spell out which is correct.)
* The tfdt box must be before the trun boxes. Moonfire NVR was not
compliant with ISO/IEC 14496-12:2015 section 8.8.12.1 before.
Chrome and Firefox didn't care, but Safari does.
* The mdat must be written with the small format. Safari is not
implementing the spec properly.
I figured these out by painstakingly comparing Moonfire NVR's output
with gpac's, making it match almost byte-for-byte until it worked, then
backing out changes one at a time to check which were relevant. Ugh!
Chrome appears to time out at 60 seconds of inactivity otherwise.
I think it's better to keep the stream open, even if the camera is
broken.
The implementation looks awkward, but that might be the state of Rust
async right now.
I spotted this by inspection: adding a media time and wall time didn't
look right. I also confirmed the brokenness on my primary NVR:
```
sqlite> .mode column
sqlite> select
...> r1.composite_id,
...> r1.prev_media_duration_90k,
...> r1.wall_duration_90k,
...> r1.media_duration_delta_90k,
...> r2.composite_id,
...> r2.prev_media_duration_90k
...> from
...> recording r1 join recording r2 on (r1.composite_id = r2.composite_id - 1)
...> where
...> r1.prev_media_duration_90k + r1.wall_duration_90k + r1.media_duration_delta_90k !=
...> r2.prev_media_duration_90k
...> limit 5;
4296791095 2232623913716 5398956 154 4296791096 2232629312672
4296791096 2232629312672 5400016 38 4296791097 2232634712688
4296791097 2232634712688 5400729 105 4296791098 2232640113417
4296791098 2232640113417 5399024 80 4296791099 2232645512441
4296791099 2232645512441 5400770 124 4296791100 2232650913211
```
In the first row, the second recording's prev_media_duration_90k is the
first's prev_media_duration_90k plus its wall time, not its media time.
The CI nightly builds had been broken with the following error:
```
error: custom inner attributes are unstable
--> /home/runner/work/moonfire-nvr/moonfire-nvr/server/target/debug/build/moonfire-db-415ce696a754c614/out/schema.rs:10:4
|
10 | #![rustfmt::skip]
| ^^^^^^^^^^^^^
|
= note: `#[deny(soft_unstable)]` on by default
= warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!
= note: for more information, see issue #64266 <https://github.com/rust-lang/rust/issues/64266>
```
I'd thought this was by mistake given that #[rustfmt::skip] is still
advertised on rustfmt's github page, but maybe not. Looks like
rust-protobuf's newest version uses
`#![cfg_attr(rustfmt, rustfmt::skip)]` to avoid this error.
Also fix a warning on nightly about an extraneous semicolon.
I also enforced some invariants in the signals code, fixing a couple
bugs. The signals code is more complex than I'd like, but hopefully
is working now.
I'd once thought about using 1 second resolution for signals and wrote
this map to match that. But I decided to match signals to the timestamps
used elsewhere instead. Match that for the days map also.
My main goal is to support creating indexes for signals as well as
recordings. An additional goal is to just shrink db.rs a bit; it's
gotten quite large.
As noted in mylog's 2b1085c:
Looks like both the GNU tools' --color argument and cargo's
CARGO_TERM_COLOR expect always/never rather than on/off. Match that.
Might as well understand off/no/false and on/yes/true also.
