mirror/ternfs-XTXMarkets
1
0
Fork 0
mirror of https://github.com/XTXMarkets/ternfs.git synced 2026-01-18 00:41:06 -06:00
Code Issues Packages Projects Releases Wiki Activity
794 Commits 5 Branches 0 Tags
27bd28ead0fae7d1708f75fcff27446c82016be9
Commit Graph

227 Commits

Author SHA1 Message Date
Francesco Mazzoli
27bd28ead0 Remove outdated comment 2023-12-10 08:39:17 +00:00
Francesco Mazzoli
788b5eed57 Fill in current block services before applying the log 
It makes a lot more sense to pick outside, given that it involves
randomness. Also, this is in preparation for shuckle picking them
in a smarter way.
 2023-12-09 15:20:24 +00:00
Francesco Mazzoli
3394328000 Do not try to close xmon fd if we don't have one 
Also, ignore errors if we can't close it. Fixes #134.
 2023-12-09 14:50:51 +00:00
Francesco Mazzoli
ab1df9137d Fix error logging when inserting stats 2023-12-08 15:57:02 +00:00
Francesco Mazzoli
128078988d Get rid of -parallel in GC 
With separate workers it's not really needed anymore.
 2023-12-08 11:51:21 +00:00
Francesco Mazzoli
5f4467d0c6 Synchronize access to in-memory block service data 
This was alread an issue before, but it never surfaced so far.
Today the quants actually hit it.
 2023-12-07 16:43:11 +00:00
Francesco Mazzoli
53049d5779 Shard batch writes, use batch UDP syscalls 
The idea is to drain the socket and do a single RocksDB WAL
write/fsync for all the write requests we have found.

The read requests are immediately executed. The reasoning here is
that currently write requests are _a lot_ slower than the read
requests because fsyncing takes ~500us on fsf1. In the future this
might change.

Since we're at it, we also use batch UDP syscalls in the CDC.

Fixes #119.
 2023-12-07 14:29:07 +00:00
Francesco Mazzoli
3eae5bbf9b Use an EMA for the in-flight CDC txns as well 2023-12-07 10:27:32 +00:00
Francesco Mazzoli
38f3d54ecd Wait forever, rather than having timeouts 
The goal here is to not have constant wakeups due to timeout. Do
not attempt to clean things up nicely before termination -- just
terminate instead. We can setup a proper termination system in
the future, I first want to see if this makes a difference.

Also, change xmon to use pipes for communication, so that it can
wait without timers as well.

Also, `write` directly for logging, so that we know the logs will
make it to the file after the logging call returns (since we now
do not have the chance to flush them afterwards).
 2023-12-07 10:11:19 +00:00
Francesco Mazzoli
af46ab2173 Bump CDC shard response timeout 2023-11-29 15:00:08 +00:00
Francesco Mazzoli
a52efe217b Tune CDC logging more 2023-11-29 14:40:33 +00:00
Francesco Mazzoli
e4c01e8728 Metrics + logging 2023-11-29 14:32:37 +00:00
Francesco Mazzoli
bd278ff6f6 Better metrics for shard responses in CDC 2023-11-29 13:52:44 +00:00
Francesco Mazzoli
4453083aa7 Correctly record request id when picking up transactions after restart 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
a367858684 Drop entire CF at once, rather than one-by-one 
A dry run of the production upgrade using a backup revealed that
dropping them one-by-one would take ages, since before we kept every
single CDC request.
 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
7537bbc6cf Remove useless line 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
fac014a864 Self-PR review, part 2 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
ba9424e224 Remove unordered_set 
Almost certainly irrelevant, but it was bugging me
 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
2eab012d76 Fix bug in poll check code 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
c94ece50cf Integer sanitizer stuff 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
59abb24a8e Add ceiling on max update size 
We don't want it to grow without bound, but we want to maximize
throughput (we'd like for fsync to not be a factor).
 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
476009381a Remove maximum enqueued requests limit 
We already drop in-flight requests that we're already processing,
so I don't think this matters very much currently.
 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
c5562c7ca3 Parallelize CDC by directory 
Fixes #66.
 2023-11-29 11:08:07 +00:00
Francesco Mazzoli
e48bb98f73 Remove outdated comment 
We now include all dependencies which are needed beyond basic build tools
 2023-11-28 21:45:40 +00:00
Francesco Mazzoli
91db9566e1 Remove option to not write out atime which is too recent 
This was pretty nasty to begin with, we now do it in the client.
 2023-11-23 13:28:23 +00:00
Francesco Mazzoli
bcf75d5308 Shut up sanitizer 2023-11-21 17:03:05 +00:00
Francesco Mazzoli
1fca8b84cd Fix type signature 2023-11-17 22:48:31 +00:00
Francesco Mazzoli
163d7b3a4d Do not return error on TIME_TOO_RECENT 
I thought we only sent it using "dontwait" for atime, but for the
normal utime calls we wait.
 2023-11-16 19:08:43 +00:00
Francesco Mazzoli
ae765b7581 Consistently check for iterator status 2023-11-16 17:12:38 +00:00
Francesco Mazzoli
b964d0632a Add option to not write out atime which is too recent 
This is to save on a ton of writes as jobs stat tons of files.
It would maybe be a bit cleaner to do it in the kmod, but this is
much quicker.

Thanks to @sgrusny for the good idea.
 2023-11-16 14:45:58 +00:00
Francesco Mazzoli
248abb2681 Fix memory leak in shards 2023-11-15 12:20:16 +00:00
Francesco Mazzoli
340e7f2f37 Harmonize addr-passing, add shuckle beacon and test it in kmod 2023-11-14 13:49:36 +00:00
Saulius Grusnys
2ce5586eb9 Periodically refresh metadata info in kmod, use two IPs for shuckle 
Fixes #112.

Co-authored-by: Francesco Mazzoli <francesco.mazzoli@xtxmarkets.com>
 2023-11-14 13:49:36 +00:00
Francesco Mazzoli
2ad278adaa Add ubuntu image to build, use jemalloc in release build 
I want to use the introspection capabilities of jemalloc, and it
should also be much faster. Preserve alpine build for go build,
it's also really useful to test inside the kmod.
 2023-11-13 15:44:55 +00:00
Francesco Mazzoli
3bc17301d6 Switch from tuple to variant for req/resp containers 
The `tuple` was for when I thought it'd be useful to leave slots
for each request, but we don't need this anymore, and now leading
up to #66 I want to be able to keep vectors of reqs/resps.
 2023-11-09 19:03:37 +00:00
Francesco Mazzoli
ad3c969772 Push full RocksDB stats to grafana 2023-11-09 16:48:51 +00:00
Francesco Mazzoli
f70c484883 Dump RocksDB full statistics to file 2023-11-09 14:12:54 +00:00
Francesco Mazzoli
057be91613 rocksDBStats -> rocksDBMetrics 2023-11-09 13:38:32 +00:00
Francesco Mazzoli
c5979a9d90 Expose some RocksDB stats 2023-11-09 13:23:49 +00:00
Francesco Mazzoli
03e9510255 Align xmon's app instances and systemd services 2023-11-08 14:36:58 +00:00
Francesco Mazzoli
ef1885a4b2 Print out more info when failing because of bad proofs 2023-11-08 11:57:32 +00:00
Francesco Mazzoli
4cc917a1c7 Expose shard socket buf size to grafana 
As a proxy to how behind shards are.
 2023-11-07 14:12:55 +00:00
Francesco Mazzoli
d0126d0656 Distinguish IO errors in eggsblocks 
See #115 for background.
 2023-11-06 19:35:05 +00:00
Francesco Mazzoli
afc4e78a62 Reduce default CDC queue size 2023-11-05 22:38:57 +00:00
Francesco Mazzoli
1ec63f9710 Implement scrubbing functionality 
Fixes #32. This also involves some reworking of the block request machinery
to make it more robust and faster. The scrubbing is done assuming that
the overwhelming majority of block checking will go through.
 2023-11-05 18:33:00 +00:00
Francesco Mazzoli
71556ce933 Switch to restech EggsFS rota 2023-11-03 14:23:44 +00:00
Francesco Mazzoli
64d400fcfe Insert shard/cdc metrics at more regular intervals 2023-11-03 13:49:38 +00:00
Francesco Mazzoli
654c0d4db4 Report CDC queue size in grafana 2023-11-03 13:49:32 +00:00
Francesco Mazzoli
674c9f22a8 Do not crash shards when swapping blocks fails 
Fixes #101
 2023-10-31 08:39:32 +00:00
Francesco Mazzoli
dd052b1919 Add excel spreadsheet to quickly adjust RocksDB size estimates 2023-10-26 14:32:35 +00:00