Most formats like mp4 require ffmpeg to make changes at the start of the file
throughout writing.
Unfortunately, this prevents us from streaming the upload as we cut it.
Instead, we spool to a temporary file until ffmpeg exits,
then upload that all at once.
In a fast cut, we edit the first and last segments then concatenate them all.
However, this leads to some tiny but perciptible artifacting around the border
of the first and second (and second-last and last) segments.
A full cut is much slower, but re-encodes the video into the desired format
and is more reliable.
We want both options to be available.
With this commit, we only add the option, we don't use it in restreamer or cutter.
To make this work, we make type a proper segment field.
We also tell get_best_segments to ignore temp segments, since they might go away
before we can actually use them.
This allows manual uploads to work without needing to fill all the edit fields
with junk.
We also set a constraint on uploader asserting that any videos from claimed onwards have a known uploader.
Again, an exception is made for DONE to allow manual uploads.
These can happen if a downloader or backfiller dies suddenly.
We treat it similarly to partial but lacking any hash.
At some point in the future we should probably have something
to find any temp segments, hash them and rename them to partials.
We wrap direct dateutil calls to handle two distinct cases:
* `common.dateutil.parse()`: We want to handle arbitrary timestamps including tz info,
then convert them to UTC.
This is used in HLS parsing, and for command line input for backfiller
* `common.dateutil.parse_utc_only()`: We want to only handle UTC timestamps,
but datetime.strptime isn't flexible enough (eg. can't handle missing fractional component).
This is used for restreamer request params.
Note this moves over the 'experimental' cutter and deletes the original cutter
that concatenates entire videos before cutting.
We may eventually want to revive that method if the experimental cutter turns out
to introduce too many issues.
We move most of the code over verbatim, but adjust it such that it acts
as a generic iterator that can be used in a variety of contexts.
Some other changes made during the move include telling ffmpeg to be quieter
(don't output version info and junk, only log if something goes wrong),
and avoiding errors during cleanup.
This is a performance optimization, allowing us to fail out early (potentially avoiding a LOT
of work) if we know we're going to reject any result that contains holes.
We add a new exception ContainsHoles that is raised in this condition.
This should help prevent changing state to EDITED with any of these fields unset,
which would blow up the cutter.
We also fix up upload_location, which was set up as a sheet input (NOT NULL DEFAULT ''),
and add a similar constraint saying any DONE columns must have non-NULL video link.
All our usage was of a single query anyway, so autocommit is easier to handle.
You can still opt into a longer transaction using the transaction() helper.
This code manages the database connections, setting their isolation level correctly
and ensuring the idempotent schema is applied before they're used.
Applying the schema on startup means we don't need to deal with the database's state,
setting it up before running, running migrations etc. However, it does put constraints on
the changes we can safely make.
Our use of seralizable isolation means that all transactions can be treated as fully
independent - the server must behave as though they'd been run seperately in some valid order.
This will give us the least surprising results when multiple connections try to modify the same
data, though we'll need to deal with occasional transaction commit failures due to conficts.
get_best_segments can sometimes take a very long time,
we don't want to stop other work from happening while it's ongoing.
So we ask gevent to run other things until there's no other work to do,
then we do one hour, then check back with gevent again.
In combination with the performance improvements, this should mean we don't block
other things from running for more than a few hundred ms at most.
strptime is much faster but can't handle as varied formats.
But in this case we fully control the format, so there's no reason not to use it.
Profiling suggests we spend about 80% of our time in get_best_segments just parsing dates,
so this is a signifigant performance gain.
The prometheus client uses a threading.Lock() to prevent shared access to
certain metric state. This lock is taken as part of doing collection, as well
as during metric.labels().
We hit a deadlock where our stack sampler signal arrived during a collection,
when the lock was held. This meant that flamegraph.labels() blocked forever,
and the lock was never released, hanging all metrics collection.
Our solution is a hack, which is to reach into the internals of our metric object
and replace its lock with a dummy one. This is reasonably safe, but only as long as
the prometheus_client internal structure doesn't change signfigiantly.
The function is quite customizable and therefore quite complex, but it allows us to
easily annotate a function to be timed with labels based on input and output,
as well as normalize results based on amount of work done to get a better
picture of the actual amount of time taken per unit of work.
This will help us monitor for performance issues.
I ran `pyflakes` on the repo and found these bugs:
```
./common/common.py:289: undefined name 'random'
./downloader/downloader/main.py:7: 'random' imported but unused
./backfiller/backfiller/main.py:150: undefined name 'variant'
./backfiller/backfiller/main.py:158: undefined name 'timedelta'
./backfiller/backfiller/main.py:171: undefined name 'sort'
./backfiller/backfiller/main.py:173: undefined name 'sort'
```
(ok, the "imported but unused" one isn't a bug, but the rest are)
This fixes those, as well as a further issue I saw with sorting of hours.
Iterables are not sortable. As an obvious example, what if your iterable was infinite?
As a result, any attempt to sort an iterable that is not already a friendly type like a list
or tuple will result in an error. We avoid this by coercing to list, fully realising the iterable
and putting it into a form that python will let us sort. It also avoids the nasty side-effect
of mutating the list that gets passed into us, which the caller may not expect. Consider this example:
```
>>> my_hours = ["one", "two", "three"]
>>> print my_hours
["one", "two", "three"]
>>> backfill_node(base_dir, node, stream, variants, hours=my_hours, order='forward')
>>> print my_hours
["one", "three", "two"]
```
Also, one of the linter errors was non-trivial to fix - we were trying to get a list of hours
(which is an api call for a particular variant), but at a time when we weren't dealing with a single
variant. My solution was to get a list of hours for ALL variants, and take the union.
This is needed by both the restreamer and the cutter, hence its inclusion in common.
The algorithm is pretty simple - it takes the 'best' segment per start time by full first,
then length of partial. All the other complexity is mainly just around detecting and reporting holes,
and being inclusive of start/end points.
Mike Lang
Mike Lang
Christopher Usher
Christopher Usher