* add docker build
* fix the working dir in Dockerfile
* add release publish docker image
* address intentation and use default release
* migrate keydb_modstatsd to keydb-internal
* rm
* add submodule cpp-statsd-client
* include trigger keydb_modstatsd Makefile in modules Makefile
* update
* have clean also trigger keydb_modstatsd clean
* move cpp-statsd-client to deps
* checkout to a06a5b9359f31d946fe163b9038586982971ae49
* update relative path in compilation
* remove submodule instead use the source
* include building keydb statsd module
* fix check in Dockerfile docker-entrypoint.sh
* fix
* fix the comment caused stuck docker build
* use commit hash as tag template
* fix
* test tag
* Revert "test tag"
This reverts commit 9cbc57137d57aab4fdd5a9283bae07391b3c7f8b.
* make docker build independent
* add new build to ci
* emit system free metrics with '/proc/meminfo'
* have emit system free memory within metrics_time_taken_us and also add metric time taken for it
* Remove Expireset (#217)
Major refactor to place expiry information directly in the object struct.
* update MemFree to MemAvailable in keydb statsd
* add metric emit for non-empty primary with less than 2 connected replicas
* address comments
* Multiply CPU percent metric by 100
* Fix memory leaks
* Fix slow to free when low lock contention
* fix nodename metricsname
* fix unnecessary replace
* Make propagating before freeing module context optional (#225)
* don't propogate on module context free for rdb load
* default in wrong place
* Flash expiration (#197)
Design Doc: https://docs.google.com/document/d/1NmnYGnHLdZp-KOUCUatX5iXpF-L3YK4VUc9Lm3Tqxpo/edit?usp=sharing
* Emit more aggregate metrics in modstatsd (#223)
* Permit keys of differing slots as long as they are served by this cluster and we are not migrating
* Fix over pessimistic checks that prevent replicas from serving mget
* Fix logic bug
* async rehash is preventing rehashing during RDB load after a db flush. Ensure it can't interefere after a flush
* make async rehash configurable
* only use internal locks when multithreaded (#205)
* Fix crossslot error migrating batches of keys
* Fix bug where we erroneously answer queries belonging to another shard
* fix mac compile
* enable level_compaction_dynamic_level_bytes after flush, and flush expires for FLASH (#229)
* enable level_compaction_dynamic_level_bytes after flush, and flush expires
* update debug reload for flash
* update debug reload for flash complete
* missing forward declare
* commit existing changes then track changes for debug reload
* missing args
* commitChanges is conditional
Co-authored-by: John Sully <jsully@snapchat.com>
---------
Co-authored-by: zliang <zliang@snapchat.com>
Co-authored-by: John Sully <jsully@snapchat.com>
Co-authored-by: Alex Cope <acope@snapchat.com>
Co-authored-by: John Sully <john@csquare.ca>
* Audit Logging for KeyProxy and KeyDB (#144)
* Audit Log: log cert fingerprint (#151)
* Add more flash storage stats to info command.
* Remove unneeded libs when not building FLASH
* Fix mem leak
* Allow the reservation of localhost connections to ensure health checks always succeed even at maxclients (#181)
* Enable a force option for commands (#183)
* Fix missing newline and excessive logging in the CLI
* Support NO ONE for "CLUSTER REPLICATE" command.
Co-authored-by: Jacob Bohac <jbohac@snapchat.com>
Co-authored-by: Sergey Kolosov <skolosov@snapchat.com>
Co-authored-by: John Sully <jsully@snapchat.com>
Co-authored-by: John Sully <john@csquare.ca>
* add hash prefixing to rocksdb
* don't need to offset internal keys
* remove hashslot prefix when loading from rocksdb
* don't prefix internal keys
* implement slot to keys with rocksdb
* fix compile
* add hashslot enum to test storage provider
* add --flash option for tests
* forgot return statement in getKeysInSlot
* check for count mismatch
* forgot ;
* add assert to ensure correctness
* fix warnings
* add option to show logs of 1 server at a time
* add fixed length prefix extractor
* add machamp
* switch machamp to main branch
* add machamp script
The change does 3 things:
1. It limits RocksDB to 256 file descriptors instead of unlimited
2. It includes the fd limit in its estimation for total file descriptors needed
3. It raises the system fd limit if possible before we open rocksdb but accounting for the 256 limit we added
Former-commit-id: 1447288209c5e7daf8a1203511fc262500ebe5e1
