## What do these changes do?
* distribute load and resource information on a heartbeat
* for each raylet, maintain total and available resource capacity as well as measure of current load
* this PR introduces a new notion of load, defined as a sum of all resource demand induced by queued ready tasks on the local raylet. This provides a heterogeneity-aware measure of load that supersedes legacy Ray's task count as a proxy for load.
* modify the scheduling policy to perform *capacity-based*, *load-aware*, *optimistically concurrent* resource allocation
* perform task spillover to the heartbeating node in response to a heartbeat, implementing heterogeneity-aware late-binding/work-stealing.
## What do these changes do?
1. Separate the log related code to logger.py from services.py.
2. Allow users to modify logging formatter in `ray start`.
## Related issue number
https://github.com/ray-project/ray/pull/2664
ray exec CLUSTER CMD [--screen] [--start] [--stop]
ray attach CLUSTER [--start]
Example:
ray exec sgd.yaml 'source activate tensorflow_p27 && cd ~/ray/python/ray/rllib && ./train.py --run=PPO --env=CartPole-v0' --screen --start --stop
This will in one command create a cluster and run the command on it in a screen session. The screen can later be attached to via ray attach. After the command finishes, the cluster workers will be terminated and the head node stopped.
This PR adds a driver table for the new GCS, which enables cleanup functionality associated with monitoring driver death.
Some testing in `monitor_test.py` is restored, but redis sharding for xray is needed to enable remaining tests.
* Fix documentation indentation.
* Add error table to GCS and push error messages through node manager.
* Add type to error data.
* Linting
* Fix failure_test bug.
* Linting.
* Enable one more test.
* Attempt to fix doc building.
* Restructuring
* Fixes
* More fixes.
* Move current_time_ms function into util.h.
* build_credis.sh: use an up-to-date credis commit.
* build_credis.sh: leveldb is updated, so update build cmds for it
* WIP: make monitor.py issue flush; switch gcs client to use credis
* Experimental: enable automatic GCS flushing with configurable policy.
* Fix linux compilation error
* Fix leveldb build
* Use optimized build for credis
* Address comments
* Attempt to fix tests
* Implement global state API for xray.
* Fix object table.
* Fixes for log structure.
* Implement cluster_resources.
* Add driver task to task table.
* Remove python flatbuffers code
* Get some global state API tests running.
* Python linting.
* Fix linting.
* Fix mock modules for doc
* Copy over flatbuffer bindings.
* Fix for tests.
* Linting
* Fix monitor crash.
* Add flake8 to Travis
* Add flake8-comprehensions
[flake8 plugin](https://github.com/adamchainz/flake8-comprehensions) that
checks for useless constructions.
* Use generators instead of lists where appropriate
A lot of the builtins can take in generators instead of lists.
This commit applies `flake8-comprehensions` to find them.
* Fix lint error
* Fix some string formatting
The rest can be fixed in another PR
* Fix compound literals syntax
This should probably be merged after #1963.
* dict() -> {}
* Use dict literal syntax
dict(...) -> {...}
* Rewrite nested dicts
* Fix hanging indent
* Add missing import
* Add missing quote
* fmt
* Add missing whitespace
* rm duplicate pip install
This is already installed in another file.
* Fix indent
* move `merge_dicts` into utils
* Bring up to date with `master`
* Add automatic syntax upgrade
* rm pyupgrade
In case users want to still use it on their own, the upgrade-syn.sh script was
left in the `.travis` dir.
* Use pep8 style
The original style file is actually just pep8 style, but with everything
spelled out. It's easier to use the `based_on_style` feature. Any overrides are
clearer that way.
* Improve yapf script
1. Do formatting in parallel
2. Lint RLlib
3. Use .style.yapf file
* Pull out expressions into variables
* Don't format rllib
* Don't allow splits in dicts
* Apply yapf
* Disallow single line if-statements
* Use arithmetic comparison
* Simplify checking for changed files
* Pull out expr into var
* Treat actor creation like a regular task.
* Small cleanups.
* Change semantics of actor resource handling.
* Bug fix.
* Minor linting
* Bug fix
* Fix jenkins test.
* Fix actor tests
* Some cleanups
* Bug fix
* Fix bug.
* Remove cached actor tasks when a driver is removed.
* Add more info to taskspec in global state API.
* Fix cyclic import bug in tune.
* Fix
* Fix linting.
* Fix linting.
* Don't schedule any tasks (especially actor creaiton tasks) on local schedulers with 0 CPUs.
* Bug fix.
* Add test for 0 CPU case
* Fix linting
* Address comments.
* Fix typos and add comment.
* Add assertion and fix test.
This adds (experimental) auto-scaling support for Ray clusters based on GCS load metrics. The auto-scaling algorithm is as follows:
Based on current (instantaneous) load information, we compute the approximate number of "used workers". This is based on the bottleneck resource, e.g. if 8/8 GPUs are used in a 8-node cluster but all the CPUs are idle, the number of used nodes is still counted as 8. This number can also be fractional.
We scale that number by 1 / target_utilization_fraction and round up to determine the target cluster size (subject to the max_workers constraint). The autoscaler control loop takes care of launching new nodes until the target cluster size is met.
When a node is idle for more than idle_timeout_minutes, we remove it from the cluster if that would not drop the cluster size below min_workers.
Note that we'll need to update the wheel in the example yaml file after this PR is merged.
* Define execution dependencies flatbuffer and add to Redis commands
* Convert TaskSpec to TaskExecutionSpec
* Add execution dependencies to Python bindings
* Submitting actor tasks uses execution dependency API instead of dummy argument
* Fix dependency getters and some cleanup for fetching missing dependencies
* C++ convention
* Make TaskExecutionSpec a C++ class
* Convert local scheduler to use TaskExecutionSpec class
* Convert some pointers to references
* Finish conversion to TaskExecutionSpec class
* fix
* Fix
* Fix memory errors?
* Cast flatbuffers GetSize to size_t
* Fixes
* add more retries in global scheduler unit test
* fix linting and cast fbb.GetSize to size_t
* Style and doc
* Fix linting and simplify from_flatbuf.
* Enable scheduling with custom resource labels.
* Fix.
* Minor fixes and ref counting fix.
* Linting
* Use .data() instead of .c_str().
* Fix linting.
* Fix ResourcesTest.testGPUIDs test by waiting for workers to start up.
* Sleep in test so that all tasks are submitted before any completes.
* Initial pass at factoring out C++ configuration into a single file.
* Expose config through Python.
* Forward declarations.
* Fixes with Python extensions
* Remove old code.
* Consistent naming for constants.
* Fixes
* Fix linting.
* More linting.
* Whitespace
* rename config -> _config.
* Move config inside a class.
* update naming convention
* Fix linting.
* More linting
* More linting.
* Add in some more constants.
* Fix linting
* WIP: removing OL, OI, TT on client exit; no saving yet.
* ray_redis_module.cc: update header comment.
* Cleanup: just the removal.
* Reformat via yapf: use pep8 style instead of google.
* Checkpoint addressing comments (partially)
* Add 'b' marker before strings (py3 compat)
* Add MonitorTest.
* Use `isort` to sort imports.
* Remove some loggings
* Fix flake8 noqa marker runtest.py
* Try to separate tests out to monitor_test.py
* Rework cleanup algorithm: correct logic
* Extend tests to cover multi-shard cases
* Add some small comments and formatting changes.
* Revert Python actor reconstruction
* Actor reconstruction using object lineage
* Add dummy arguments and return values for actor tasks
* Pin dummy outputs for actor tasks
* Skip checkpointing test for now
* TODOs
* minor edits
* Generate dummy object dependencies in Python, not C
* Fix linting.
* Move actor counter and dummy objects inside of the actor handle
* Refactor Worker._process_task, suppress exception propagation for
sequential actor tasks
* Reconstruct actor state when local schedulers fail.
* Simplify construction of arguments to pass into default_worker.py from local scheduler.
* Remove deprecated ray.actor.
* Simplify actor reconstruction method.
* Fix linting.
* Small fixes.
* 4 space indentation for actor.py.
* 4 space indentation for worker.py.
* 4 space indentation for more files.
* 4 space indentation for some test files.
* Check indentation in Travis.
* 4 space indentation for some rl files.
* Fix failure test.
* Fix multi_node_test.
* 4 space indentation for more files.
* 4 space indentation for remaining files.
* Fixes.
* Implement sharding in the Ray core
* Single node Python modifications to do sharding
* Do the sharding in redis.cc
* Pipe num_redis_shards through start_ray.py and worker.py.
* Use multiple redis shards in multinode tests.
* first steps for sharding ray.global_state
* Fix problem in multinode docker test.
* fix runtest.py
* fix some tests
* fix redis shard startup
* fix redis sharding
* fix
* fix bug introduced by the map-iterator being consumed
* fix sharding bug
* shard event table
* update number of Redis clients to be 64K
* Fix object table tests by flushing shards in between unit tests
* Fix local scheduler tests
* Documentation
* Register shard locations in the primary shard
* Add plasma unit tests back to build
* lint
* lint and fix build
* Fix
* Address Robert's comments
* Refactor start_ray_processes to start Redis shard
* lint
* Fix global scheduler python tests
* Fix redis module test
* Fix plasma test
* Fix component failure test
* Fix local scheduler test
* Fix runtest.py
* Fix global scheduler test for python3
* Fix task_table_test_and_update bug, from actor task table submission race
* Fix jenkins tests.
* Retry Redis shard connections
* Fix test cases
* Convert database clients to DBClient struct
* Fix race condition when subscribing to db client table
* Remove unused lines, add APITest for sharded Ray
* Fix
* Fix memory leak
* Suppress ReconstructionTests output
* Suppress output for APITestSharded
* Reissue task table add/update commands if initial command does not publish to any subscribers.
* fix
* Fix linting.
* fix tests
* fix linting
* fix python test
* fix linting
* Change local scheduler bookkeeping to use GPU IDs.
* Update actor test.
* Add tests for actors and tasks simultaneously using GPUs.
* Add additional task GPU ID test.
* Fix linting.
* Make redis GPU assignment ignore GPU IDs.
* Small fix.
* Clean up state when drivers exit.
* Remove unnecessary field in ActorMapEntry struct.
* Have monitor release GPU resources in Redis when driver exits.
* Enable multiple drivers in multi-node tests and test driver cleanup.
* Make redis GPU allocation a redis transaction and small cleanups.
* Fix multi-node test.
* Small cleanups.
* Make global scheduler take node_ip_address so it appears in the right place in the client table.
* Cleanups.
* Fix linting and cleanups in local scheduler.
* Fix removed_driver_test.
* Fix bug related to vector -> list.
* Fix linting.
* Cleanup.
* Fix multi node tests.
* Fix jenkins tests.
* Add another multi node test with many drivers.
* Fix linting.
* Make the actor creation notification a flatbuffer message.
* Revert "Make the actor creation notification a flatbuffer message."
This reverts commit af99099c8084dbf9177fb4e34c0c9b1a12c78f39.
* Add comment explaining flatbuffer problems.
* Ignore deleted clients when reading address info from Redis
* Remove self from db_client table when exiting cleanly
* Fix valgrind test
* Do not call plasma_perform_release when disconnecting
* Failing test case
* Local scheduler exits cleanly after plasma store dies
* Tolerate one plasma store failure
* Tolerate plasma store failures on all nodes except head node
* Plasma manager heartbeats
* Component failure tests
* Don't run the helper for Python testing
* Fix C test
* Fix hanging plasma transfer test
* Fix python3
* Consolidate ClientConnection code
* Fix valgrind test
* fix c test
* We can restart worker nodes!
* Fix flatbuffers bug
* Address comments
* Only register actual workers with the local scheduler
* Fix bug
* Fix segfaults
* Add test case that tests for driver liveness, fix local scheduler bug
* Clean up after tests
* Allocate retry info on the stack
* Send SIGKILL before waiting
* Relax unit test conditions
* Driver liveness test case and documentation
* Compile the Ray redis module with C++.
* Redo parsing of object table notifications with flatbuffers.
* Update redis module python tests.
* Redo parsing of task table notifications with flatbuffers.
* Fix linting.
* Redo parsing of db client notifications with flatbuffers.
* Redo publishing of local scheduler heartbeats with flatbuffers.
* Fix linting.
* Remove usage of fixed-width formatting of scheduling state in channel name.
* Reply with flatbuffer object to task table queries, also simplify redis string to flatbuffer string conversion.
* Fix linting and tests.
* fix
* cleanup
* simplify logic in ReplyWithTask
* Clean up plasma subscribers on EPIPE
First pass at a monitoring script - monitor can detect local scheduler death
Clean up task table upon local scheduler death in monitoring script
Don't schedule to dead local schedulers in global scheduler
Have global scheduler update the db clients table, monitor script cleans up state
Documentation
Monitor script should scan tables before beginning to read from subscription channel
Fix for python3
Redirect monitor output to redis logs, fix hanging in multinode tests
* Publish auxiliary addresses as part of db_client deletion notifications
* Fix test case?
* Small changes.
* Use SCAN instead of KEYS
* Address comments
* Address more comments
* Free redis module strings
Alexey Tumanov