Skipping these makes the sequence numbering inconcistent; we've received
a file and suppsedly added it to the database, but if you check the
sequence number afterwards it didn't increase, i.e., we trigger [this
failure
condition](47f48faed7/lib/model/indexhandler.go (L447-L459))
and, similarly, a future update will look like there was a hole in the
numbering.
I propose to at least temporarily remove this optimisation in order for
things to make more sense. Is there a reason to keep this beyond saving
some database operations?
### Purpose
Resend our indexes since we fixed that index-sending issue.
I made a new thing to only drop the non-local-device index IDs, i.e.,
those for other devices. This means we will see a mismatch and resend
all indexes, but they will not. This is somewhat cleaner as it avoids
resending everything twice when two devices are upgraded, and in any
case, we have no reason to force a resend of incoming indexes here.
### Testing
It happens on my computer...
This changes the GC mechanism so that the first pass (which reads all
FileInfos to populate bloom filters with block & version hashes) can
happen concurrently with normal database operations.
The big gcMut still exists, and we grab it temporarily to block all
other modifications while we set up the bloom filters. We then release
the lock and let other things happen, with those other things also
updating the bloom filters as required. Once the first phase is done we
again grab the gcMut, knowing that we are the sole modifier of the
database, and do the cleanup.
I also removed the final compaction step.
If the GC finds a key k that it wants to keep, it records that in a
Bloom filter. If a key k' can be removed but its hash collides with k,
it will be kept. Since the old Bloom filter code was completely
deterministic, the next run would encounter the same collision, assuming
k must still be kept.
A randomized hash function that uses all the SHA-256 bits solves this
problem: the second run has a non-zero probability of removing k', as
long as the Bloom filter is not completely full.
Group the global list of files by version, instead of having one flat list for all devices. This removes lots of duplicate protocol.Vectors.
Co-authored-by: Jakob Borg <jakob@kastelo.net>
This adds indirection of large version vectors in the same manner as we
already to block lists. The effect is the same: less duplicated data in
some situations.
To mitigate the impact for when this indirection
wouldn't be needed I've added an indirection cutoff for both blocks and
the new version vector stuff: we don't do the indirection at all for
small block lists or small version vectors, instead storing it directly
like we used to do. This is faster for small files and small setups.
This makes the GC runner a service that will stop fairly quickly when
told to.
As a bonus, STTRACE=app will print the service tree on the way out,
including any errors they've flagged.