161sh/CatgirlIntelligenceAgency
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
1,245 Commits 19 Branches 14 Tags 14 MiB
abbadc92a0
Commit Graph

1245 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Viktor Lofgren
064bc5ee76 (processed-data) New parquet-serializable models for converter output 2023-09-11 14:08:40 +02:00
Viktor Lofgren
a52d78c8ee (work-log) New batching work log 2023-09-11 14:08:08 +02:00
Viktor Lofgren
a00cabe223 (parquet-floor) Patch in support for writing and reading repeated values 2023-09-11 14:06:43 +02:00
Viktor Lofgren
dbe974f510 (parquet) Use ZSTD compression by default. 2023-09-11 09:02:58 +02:00
Viktor Lofgren
a284682deb (parquet) Add parquet library 
This small library, while great, will require some modifications to fit the project's needs, so it goes into third-party directly.
 2023-09-05 10:38:51 +02:00
Viktor Lofgren
07d7507ac6 (control-service) Move Actions up in storage-details 
Papercut fix. If a file storage area has a lot of files, you have to scroll down a long way to get to the actions otherwise.
 2023-09-02 15:41:55 +02:00
Viktor Lofgren
c68d17d482 (keyword-extraction) Fix bug leading to position data missing on some keywords. 
This was due to a discrepancy between the KeywordPositionBitmask and WordsTfIdfCounts' concept of a keyword.
 2023-09-02 14:48:55 +02:00
Viktor Lofgren
9e185e80ce (control-service) Add timestamp to file storages. 2023-09-02 14:01:04 +02:00
Viktor Lofgren
676e7c7947 (keywords) Add Serializable properties that went missing as the record became a class 2023-09-02 09:52:01 +02:00
Viktor Lofgren
04212b2cef (btree) Add more consistent asserts on sortedness 2023-09-01 15:45:02 +02:00
Viktor Lofgren
bafc2a1f30 (reverse-index) Force() final docs after being written 
Unlikely to be a problem, but we want to ensure it's on dsik before we go read it later.
 2023-09-01 15:43:53 +02:00
Viktor Lofgren
563e388a45 (reverse-index) Fix parallel documents sorting bug 
Bug was caused by parallel sorting capturing the iterator rather than the offsets to sort.
 2023-09-01 15:42:45 +02:00
Viktor Lofgren
d31d8ec5b0 (index) Log keyword ids on hex format 2023-09-01 15:40:24 +02:00
Viktor Lofgren
2b00cd632d (process) Propagate environment JVM params to the index constructor 2023-09-01 15:39:42 +02:00
Viktor Lofgren
5f427d2b4c (keywords) Clean up leaky abstractions, clean up tests 2023-09-01 13:52:00 +02:00
Viktor Lofgren
8c0ce4fc1d (index journal; minor) Clean up 2023-09-01 11:32:24 +02:00
Viktor Lofgren
10a74f45ea (index journal; minor) Even cleaner separation of concerns. 2023-09-01 11:28:02 +02:00
Viktor Lofgren
320dad7f1a (index journal) Fix leaky abstraction in IndexJournalReader. 
The caller shouldn't be required to know the on-disk layout of the file to make use of the data in a performant way.
 2023-09-01 11:18:13 +02:00
Viktor Lofgren
88ac72c8eb (journal/reverse index) Working WIP fix over-allocation of documents 2023-08-31 20:16:02 +02:00
Viktor Lofgren
f74b9df0a7 (array) Don't use paging arrays when mapping small files for writing 2023-08-31 20:15:10 +02:00
Viktor Lofgren
a6f1335375 (loader) Fix bugfix where the loader would omit some meta and words. 2023-08-31 17:48:43 +02:00
Viktor Lofgren
f321fa5ad3 (array) Override to Paging...Array$range() 
This is a big performance boost in array.range().get().

Without an override, each access will go through pages[page].get(...) for each get()-operation.  This adds up very quickly.  BTreeReader does a bunch of get():s on a range()'d array during traversal in the queryData... methods.
 2023-08-31 13:52:29 +02:00
Viktor Lofgren
03d999444d (ldb) Re-add accidentally removed stmt.addBatch that breaks 2023-08-31 12:06:30 +02:00
Viktor Lofgren
763ed260c3 (ldb) Better handling of null pubYear 2023-08-30 23:08:27 +02:00
Viktor Lofgren
764e7d1315 (index) Add more comprehensive integration tests for the index service. 2023-08-30 10:37:24 +02:00
Viktor Lofgren
048f685073 (ldb) add OR IGNORE to insert status query 
Otherwise it will sometimes fail because documents may appear more than once in error scenarios.
 2023-08-30 10:34:01 +02:00
Viktor Lofgren
e4d7958379 (control) ProcessLivenessMonitorActor shouldn't reap tasks based on service instance liveness 2023-08-29 18:19:04 +02:00
Viktor
bdcbfb11a8
Merge pull request #42 from MarginaliaSearch/no-downtime-upgrades 
Zero downtime upgrades, merge-based index construction
 2023-08-29 17:05:48 +02:00
Viktor Lofgren
3f288e264b (minor) Clean up dead endpoints 2023-08-29 17:04:54 +02:00
Viktor Lofgren
dd593c292c (loader) Minor optimizations and bugfixes. 
* Reduce memory churn in LoaderIndexJournalWriter, fix bug with keyword mappings as well
* Remove remains of OldDomains
* Ensure LOADER_PROCESS_OPTS gets fed to the processes
* LinkdbStatusWriter won't execute batch after each added item post 100 items
 2023-08-29 15:37:52 +02:00
Viktor Lofgren
fa87c7e1b7 (process) Automatic flightrecorder runs for processes when run in docker. 2023-08-29 14:12:51 +02:00
Viktor Lofgren
39c1857c61 (heartbeat, reverse-index) Better heartbeat mocking, improved heartbeats for reverse index construction. 2023-08-29 13:07:55 +02:00
Viktor Lofgren
c57a2d0dc3 (control-service) Remove old index journal files when restoring a backup. 2023-08-29 11:58:01 +02:00
Viktor Lofgren
a2e6616100 (index-reverse) Add documentation and clean up code. 2023-08-29 11:35:54 +02:00
Viktor Lofgren
ba4513e82c (loader) Revert accidental experimental changes that slipped by in an earlier commit 2023-08-28 19:54:56 +02:00
Viktor Lofgren
6525b16e1f (minor) Improved logging and error messages 2023-08-28 19:53:55 +02:00
Viktor Lofgren
b6a92506d1 (index) Hook in missing DocIdRewriter 
This enables documents to be ranked properly.
 2023-08-28 19:53:43 +02:00
Viktor Lofgren
ffa0366deb (minor) Fix typo in ActorStateMachine's logging 2023-08-28 16:11:52 +02:00
Viktor Lofgren
00c4686ef0 (reverse-index) Fix over-allocation of the count array in merging 2023-08-28 14:36:28 +02:00
Viktor Lofgren
3101b74580 (index) Move to a lexicon-free index design 
This is a system-wide change.  The index used to have a lexicon, mapping words to wordIds using a large in-memory hash table.   This made index-construction easier, but it
also added a fairly significant RAM penalty to both the index service and the loader.

The new design moves to 64 bit word identifiers calculated using the murmur hash of the keyword, and an index construction based on merging smaller indices.

It also became necessary half-way through to upgrade guice as its error reporting wasn't *quite* compatible with JDK20.
 2023-08-28 14:02:23 +02:00
Viktor Lofgren
4e694fdff6 (minor) Comment build.gradle 2023-08-25 16:40:53 +02:00
Viktor Lofgren
194a6057dd (index,control) Recoverable index backups 2023-08-25 14:57:43 +02:00
Viktor Lofgren
e710e057e2 (db) Remove EC_URL and EC_PAGE_DATA from mariadb database 2023-08-25 13:45:03 +02:00
Viktor Lofgren
28188a6e59 (control) Simplify ConvertAndLoadActor 2023-08-25 13:30:20 +02:00
Viktor Lofgren
70a5df96c8 (control) Display progress of process tasks 2023-08-25 13:05:21 +02:00
Viktor Lofgren
460998d512 (index) Move index construction to separate process. 
This provides a much cleaner separation of concerns, and makes it possible to get rid of a lot of the gunkier parts of the index service.  It will also permit lowering the Xmx on the index service a fair bit, so we can get CompressedOOps again :D
 2023-08-25 12:52:54 +02:00
Viktor Lofgren
e741301417 (search) Remove endpoint flush-search-caches 
It's not necessary anymore with the new linkdb.
 2023-08-25 09:51:06 +02:00
Viktor Lofgren
5ed5298409 (converter) Update confusing state description 
SWAP_LEXICON doesn't instruct the index service to do anything.  It just moves the file.
 2023-08-24 18:56:49 +02:00
Viktor Lofgren
b911665691 (index) Clean up and optimize valuator 2023-08-24 18:34:06 +02:00
Viktor Lofgren
56eb83319d (index) Clean up result domain deduplicator 2023-08-24 18:24:55 +02:00