hiro/ray
1
0
Fork 0
mirror of https://github.com/vale981/ray synced 2025-03-04 17:41:43 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

config

Browse source 
Signed-off-by: Clarence Ng <clarence.wyng@gmail.com>
This commit is contained in:
Clarence Ng 2022-08-12 16:16:32 -07:00
parent dff0385ed5
commit 62c123b352
1 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
doc/source/ray-core/scheduling/configure-memory-monitor.rst
View file

@ -17,11 +17,11 @@ When the memory monitor detects the usage is above the threshold, the raylet wil
Configuration parameters
------------------------
The memory monitor is disabled by default and is enabled via environment variables.
The memory monitor is configured via environment variables.
`RAY_memory_usage_threshold_fraction [default_value=0, min=0.0, max=1.0, recommended=0.9]`: The usage fraction is calculated as (node memory used / node memory available). The memory here includes the shared memory allocated for the object store.
`RAY_memory_usage_threshold_fraction [default_value=0.95, min=0.0, max=1.0]`: The usage fraction is calculated as (node memory used / node memory available). The memory here includes the shared memory allocated for the object store.
`RAY_memory_monitor_interval_ms [default_value=0, recommended=500]`: The monitor is disabled when this value is 0. This is the frequency in which the monitor computes the memory usage and notifies the raylet if the usage exceeds the threshold. Running the monitor more frequently allows it to respond faster to memory growth.
`RAY_memory_monitor_interval_ms [default_value=200]`: The monitor is disabled when this value is 0. This is the frequency in which the monitor computes the memory usage and notifies the raylet if the usage exceeds the threshold. Running the monitor more frequently allows it to respond faster to memory growth.
Detect and reduce worker evictions
-----------------------------------