Backend-ai-manager

Latest version: v1.3.5

Safety actively analyzes 688705 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 2 of 7

1.2.2

Not secure
------------------

- Add metadata/aliases for TensorFlow v1.5 kernel images to the default sample configs.

- Polish CI and test suites.

- Add etcd put/get/del manager CLI commands to get rid of the necessity of an extra
etcdcli binary during installation. (lablup/backend.ai15)

1.2.1

Not secure
------------------

- Minor update to fix dependency versions.

1.2.0

Not secure
------------------

**NOTICE**

- From this release, the manager and agent versions will go together, which indicates
the compatibility of them, even when either one has relatively little improvements.

**CHANGES**

- The gateway server now consider per-agent image availability when scheduling a new
kernel. (29)

- The execute API now returns exitCode value of underlying subprocesses in the batch
mode. (60)

- The gateway server is now fully horizontally-scalable.
There is no states shared via multiprocessing shared memory and all such states are
now managed by a separate Redis instance.

- Improve logging: it now provides multiprocess-safe file-based rotating logs. (10)

- Fix the Admin API error when filtering agents by their status due to a missing
method parameter in ``Agent.batch_load()``.

1.1.0

Not secure
------------------

**NOTICE**

- Requires alembic database migration for upgrading.

**API CHANGES**

- The semantic for client session token changes. (56, 58)
Clients may reuse the same session token across different sessions if only a single
session is running at a time.
The manager now returns an explicit error if the client request is going to violate
this constraint.

- In the API responses, Rate-Limit-Reset is gone away and now we have
Rate-Limit-Window value instead. (55)

Since we use a rolling counter, there is no explicit reset point but you are now
guaranteed to send at most N requests for the last 15 minutes (where N is the
per-user rate limit) at ANY moment.

- When continuing or sending user-inputs via the execute API, you
must set the mode field to "continue" or "input" respectively.

- You no longer have to specify a random run ID on the first request of a run during
session; if the field is set to null, the server will assign a new run ID
automatically. Note that you STILL have to specify the run ID on subsequent
requests for the run. (59)

All API responses now include its corresponding run ID regardless of whether it is
given by the client or assigned by the server, which eases client-side
demultiplexing of concurrent executions.

**OTHER IMPROVEMENTS**

- Fix atomicity of rate-limiting calculation in multi-core setups. (55)

- Remove simplejson from dependencies in favor of the standard library.
The stdlib has been updated to support all required features and use
an internal C-based module for performance.

1.0.4

Not secure
------------------

- Minor update for execute API: allow explicit continue/input mode values.

- Mitigate connection failures after a DB failover event. (35)

1.0.3

Not secure
------------------

- Add virtual folder!

- Update aioredis to v1.0.0 release.

- Remove "mode" argument when calling agent RPC "get completions" calls.

Page 2 of 7

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.