Backend.ai

Fixes
* Fix broken wheel build workflow due to cached build packages by updating them ([219](https://github.com/lablup/backend.ai-client-py/issues/219))

Miscellaneous
* Explicitly show an warning log upon timeout of `GlobalTimer` acquiring a distributed lock and increase the default lock acquire timeout to 9600 seconds ([138](https://github.com/lablup/backend.ai-common/issues/138))

Features
* Let `validators.TimeDuration` to accept numbers (int and float) as well ([128](https://github.com/lablup/backend.ai-common/issues/128))
* Migrate `manager.distribute` and make it to use pluggable distributed lock implementations ([129](https://github.com/lablup/backend.ai-common/issues/129))
* Allow `EventDispatcher` and `EventProducer` to have a custom stream key other than "events" ([131](https://github.com/lablup/backend.ai-common/issues/131))
* Add `EtcdLock`, a locking mechanism based on Etcd's lock feature. ([133](https://github.com/lablup/backend.ai-common/issues/133))
* Add lifetime to the distributed locks to ensure the liveness property (i.e., guaranteed automatic release of locks in case of hangs or network partitions) ([136](https://github.com/lablup/backend.ai-common/issues/136))

Fixes
* Prevent a potential deadlock when all executor threads is occupied by the lock acquiring methods and unlocking via the executor becomes impossible ([132](https://github.com/lablup/backend.ai-common/issues/132))
* Migrate the bgtask framework from the manager and make it reusable in other projects such as storage-proxy ([135](https://github.com/lablup/backend.ai-common/issues/135))

Miscellaneous
* Remove old codes in ImageRef class, which was used to perform Etcd operations. ([127](https://github.com/lablup/backend.ai-common/issues/127))
* Run the redis test suite only when modules under common.redis or their test codes are changed ([130](https://github.com/lablup/backend.ai-common/issues/130))

Features
* Replace `aetcd3` library with `etcetra`. ([124](https://github.com/lablup/backend.ai-common/issues/124))
* Accept async pipeline builders in `redis.execute()` instead of silently failing by returning pipeline objects ([125](https://github.com/lablup/backend.ai-common/issues/125))

Fixes
* Fix argument typing of `redis.execute_script()` ([126](https://github.com/lablup/backend.ai-common/issues/126))

Fixes
* Remove duplicated logic to obtain redis local time in streaming proxy requests, which caused the depletion of the Redis connection pool. ([576](https://github.com/lablup/backend.ai-manager/issues/576))
* Fix regression of the logic in getting the container stats per period. ([577](https://github.com/lablup/backend.ai-manager/issues/577))
* Make updating scaling group works again by converting the primitive values of the `scheduler_opts` to a `ScalingGroupsOpts` type to pass the trafaret validation step. ([578](https://github.com/lablup/backend.ai-manager/issues/578))
* Remove explicit usage of `pg_advisory_lock` on scheduler. ([579](https://github.com/lablup/backend.ai-manager/issues/579))

Features
* Allow specifying the architecture when setting image aliases ([211](https://github.com/lablup/backend.ai-client-py/issues/211))
* Add `callback_url` parameter in session creation API/CLI to support session event webhook. ([217](https://github.com/lablup/backend.ai-client-py/issues/217))

Fixes
* Remove invalid "uuid" field from `groups` model. ([216](https://github.com/lablup/backend.ai-client-py/issues/216))
* Remove access key option in vfolder list command since manager does not support the field. ([218](https://github.com/lablup/backend.ai-client-py/issues/218))