[
https://issues.apache.org/jira/browse/IMPALA-9989?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17185373#comment-17185373
]
Qifan Chen commented on IMPALA-9989:
------------------------------------
Some work is done to address the current limitation in admission controller by
appending the last known memory consumption statistics about a pool
to the existing memory exhaustion message. The statistics is
logged in impalad.INFO when a query is queued or timed out due to
memory pressure in the pool or on the host. The statistics can also be
part of the query profile.
The BNF of the new memory consumption statistics is as follows.
topN_query_stats ::=
queries: a list of query Ids and memory consumed for up to 5 queries
with top memory consumptions
total_consumed: total memory consumed by these topN queries
fraction_of_pool_total_mem: total memory consumed divided
by pool memory usage (if
feasible to report)
all_query_stats ::=
num_running: the total number of queries running
min: the minimal memory consumption of all running queries
max: the maximal memory consumption of all running queries
pool_total_mem: the total memory consumption of all running queries
average: the average memory consumption of all running queries
(if feasible to report)
pool_stats ::=
<pool_name> ":"
<topN_query_stats>
<all_query_stats>
stats_on_host ::=
"Stats for host " <host>
List of <pool_stats>
aggregated_pool_stats ::=
"Aggregated stats for pool " <pool_name>
<topN_query_stats>
memory_consumption_statistics ::=
<stats_on_host> | <aggregated_pool_stats>
The stats_on_host describes memory consumption for every pool on
a host and is useful in analyzing memory exhaustion on that host.
The aggregated_pool_stats describes the aggregated memory consumption
on all hosts for a pool for a set of queries and is useful in analyzing
memory exhaustion in that pool.
Example of stats_on_host for pool root.queueB and root.queueC on
host host1:25000.
Stats for host host1:25000
pool_name=root.queueB:
topN_query_stats:
topN_query_stats:
queries=[
id=0000000000000001:0000000000000004, consumed=20.00 MB,
id=0000000000000001:0000000000000003, consumed=19.00 MB,
id=0000000000000001:0000000000000002, consumed=8.00 MB
],
total_consumed=47.00 MB
fraction_of_pool_total_mem=0.47
all_query_stats:
num_running=4,
min=5.00 MB,
max=20.00 MB,
pool_total_mem=100.00 MB,
average=25.00 MB
pool_name=root.queueC:
topN_query_stats:
queries=[
id=0000000000000002:0000000000000000, consumed=18.00 MB,
id=0000000000000002:0000000000000001, consumed=12.00 MB
],
total_consumed=30.00 MB
fraction_of_pool_total_mem=0.06
all_query_stats:
num_running=40,
min=10.00 MB,
max=200.00 MB,
pool_total_mem=500.00 MB,
average=12.50 MB
Aggregated stats for pool root.queueC:
topN_query_stats:
queries=[
id=0000000000000002:0000000000000001, consumed=32.00 MB,
id=0000000000000002:0000000000000004, consumed=26.00 MB,
id=0000000000000002:0000000000000000, consumed=21.00 MB,
id=0000000000000002:0000000000000002, consumed=17.00 MB,
id=0000000000000002:000000000000000e, consumed=9.00 MB
],
total_consumed=105.00 MB
fraction_of_pool_total_mem=0.82
> Improve admission control pool stats logging
> --------------------------------------------
>
> Key: IMPALA-9989
> URL: https://issues.apache.org/jira/browse/IMPALA-9989
> Project: IMPALA
> Issue Type: Improvement
> Reporter: Vincent Tran
> Assignee: Qifan Chen
> Priority: Major
>
> Information that should be explicit to log consumers:
> 1) Global pool stats at the time of admission. The stats from
> 'admission-controller.cc:515' only aggregate from queries admitted by this
> host.
> 2) Local host's memory - since it is also a factor in the admission decision.
> 3) Any other info that would factor into the admission decision.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
