cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexey Plotnik (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (CASSANDRA-5552) Cassandra doesn't preload key caches from disk on start
Date Thu, 09 May 2013 13:43:16 GMT

     [ https://issues.apache.org/jira/browse/CASSANDRA-5552?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Alexey Plotnik updated CASSANDRA-5552:
--------------------------------------

    Description: 
My cache related settings is default:
{quote}
key_cache_size_in_mb: # no value here
key_cache_save_period: 14400
row_cache_size_in_mb: 0
row_cache_save_period: 0
row_cache_provider: SerializingCacheProvider
saved_caches_directory: /var/opt/cassandra/data/saved_caches
{quote}

Every time I restart Cassandra node with *service cassandra stop; service cassandra start*
commands I see with Zabbix or JMX or *nodetool info* the key cache has been nulled. It happens
every time I restart the node. This situation is stable on all the nodes of the cluster. I
guessed since the key cache is flushed to disk every 4 hours I should never see the situation
with cold caches.

The output of *nodetool info* after 30 minutes from node has been started:
bq.Key Cache        : size 2105664 (bytes), capacity 104857584 (bytes), 3812 hits, 47610 requests,
0.371 recent hit rate, 14400 save period in seconds

Before restarting it looked as follows:
bq.Key Cache        : size 104857584 (bytes), capacity 104857584 (bytes), 79777150 hits, 502884072
requests, 0.847 recent hit rate, 14400 save period in seconds

All the nodes have the cache capacity auto-calculated to 104857584 (100 Mb) according to settings
I put above.

The capacity of saved_caches directory on all the nodes is strange: with command

bq.ansible cassandra -i hosts -m raw -a 'du -s /var/opt/cassandra/data/saved_caches' | grep
caches
I see values from 50 to 100 Kb only, against the values of 100Mb the `nodetool info` outputs.

After node restart in 1 week the key cache is filled up to ~2,2 million of keys, but when
I perform node restart (after running for few days) it is nulled again.

I've never changed the default configuration of cassandra caches. What's going wrong? Thanks.

  was:
My cache related settings is default:
{quote}
key_cache_size_in_mb: # no value here
key_cache_save_period: 14400
row_cache_size_in_mb: 0
row_cache_save_period: 0
row_cache_provider: SerializingCacheProvider
saved_caches_directory: /var/opt/cassandra/data/saved_caches
{quote}

Every time I restart Cassandra node with *service cassandra stop; service cassandra start*
I see with Zabbix or JMX or *nodetool info* the key cache was nulled. It happens every time
I restart the node. This situation is on all the nodes I have. I guessed since key cache is
flushed to disk every 4 hours I should never see the situation with cold caches.

nodetool info output after 30 minutes from node has been started:
bq.Key Cache        : size 2105664 (bytes), capacity 104857584 (bytes), 3812 hits, 47610 requests,
0.371 recent hit rate, 14400 save period in seconds

Before restarting it looked as follows:
bq.Key Cache        : size 104857584 (bytes), capacity 104857584 (bytes), 79777150 hits, 502884072
requests, 0.847 recent hit rate, 14400 save period in seconds

All the nodes have cache capacity auto-calculated to 104857584 (100 Mb)

The capacity of saved_caches directory on all the nodes is strange: with

bq.ansible cassandra -i hosts -m raw -a 'du -s /var/opt/cassandra/data/saved_caches' | grep
caches
I see values from 50-100 Kb only against values of 100Mb the `nodetool info` outputs.

After node has been restarted in 1 week key cache is filled up to ~2,2 million of keys, but
when I perform node restart (after > 4 hours of running) it nulled again.

I've never changed the default configuration of cassandra caches.

    
> Cassandra doesn't preload key caches from disk on start
> -------------------------------------------------------
>
>                 Key: CASSANDRA-5552
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-5552
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 1.1.9
>         Environment: CentOS 6.2, Datastax Cassandra Commutnity Edition package
>            Reporter: Alexey Plotnik
>              Labels: performance
>
> My cache related settings is default:
> {quote}
> key_cache_size_in_mb: # no value here
> key_cache_save_period: 14400
> row_cache_size_in_mb: 0
> row_cache_save_period: 0
> row_cache_provider: SerializingCacheProvider
> saved_caches_directory: /var/opt/cassandra/data/saved_caches
> {quote}
> Every time I restart Cassandra node with *service cassandra stop; service cassandra start*
commands I see with Zabbix or JMX or *nodetool info* the key cache has been nulled. It happens
every time I restart the node. This situation is stable on all the nodes of the cluster. I
guessed since the key cache is flushed to disk every 4 hours I should never see the situation
with cold caches.
> The output of *nodetool info* after 30 minutes from node has been started:
> bq.Key Cache        : size 2105664 (bytes), capacity 104857584 (bytes), 3812 hits, 47610
requests, 0.371 recent hit rate, 14400 save period in seconds
> Before restarting it looked as follows:
> bq.Key Cache        : size 104857584 (bytes), capacity 104857584 (bytes), 79777150 hits,
502884072 requests, 0.847 recent hit rate, 14400 save period in seconds
> All the nodes have the cache capacity auto-calculated to 104857584 (100 Mb) according
to settings I put above.
> The capacity of saved_caches directory on all the nodes is strange: with command
> bq.ansible cassandra -i hosts -m raw -a 'du -s /var/opt/cassandra/data/saved_caches'
| grep caches
> I see values from 50 to 100 Kb only, against the values of 100Mb the `nodetool info`
outputs.
> After node restart in 1 week the key cache is filled up to ~2,2 million of keys, but
when I perform node restart (after running for few days) it is nulled again.
> I've never changed the default configuration of cassandra caches. What's going wrong?
Thanks.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message