mirror of
https://github.com/vale981/ray
synced 2025-03-06 18:41:40 -05:00
cffe8f9806
* wip * wip * format * wip * note * lint * fix * flag * typo * raise timeout * fix * optional get * fix flag * increase timeout in test * update docs * format
15 lines
723 B
ReStructuredText
15 lines
723 B
ReStructuredText
Redis Memory Management (Experimental)
|
|
======================================
|
|
|
|
Ray stores metadata associated with tasks and objects in one or more Redis
|
|
servers, as described in `An Overview of the Internals
|
|
<internals-overview.html>`_. Applications that are long-running or have high
|
|
task/object generation rate could risk high memory pressure, potentially leading
|
|
to out-of-memory (OOM) errors.
|
|
|
|
In Ray `0.6.1+` Redis shards can be configured to LRU evict task and object
|
|
metadata by setting ``redis_max_memory`` when starting Ray. This supercedes the
|
|
previously documented flushing functionality.
|
|
|
|
Note that profiling is disabled when ``redis_max_memory`` is set. This is because
|
|
profiling data cannot be LRU evicted.
|