[
https://issues.apache.org/jira/browse/CASSANDRA-20162?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitry Konstantinov updated CASSANDRA-20162:
--------------------------------------------
Description:
Currently for memtable_allocation_type: offheap_objects we allocate memory to
calculate dataSize() during a write to a memtable because we use default
implementation of dataSize() which retrieves an object using value() method.
To avoid such memory allocation we can calculate data size directly for
NativeCell and NativeClustering
test schema details:
{code:java}
CREATE TABLE memory_tests.test_table_text_0 (part_key text,clust_key
text,value_0 text,value_1 text,value_2 text,value_3 text,value_4 text,value_5
text,value_6 text,value_7 text,value_8 text,value_9 text,PRIMARY
KEY(part_key,clust_key))
{code}
write query:
{code:java}
INSERT INTO memory_tests.test_table_text_0
(part_key,clust_key,value_0,value_1,value_2,value_3,value_4,value_5,value_6,value_7,value_8,value_9)
VALUES (?,?,?,?,?,?,?,?,?,?,?,?)
{code}
changed settings:
{code:java}
memtable_allocation_type: offheap_objects
memtable:
configurations:
skiplist:
class_name: SkipListMemtable
trie:
class_name: TrieMemtable
default:
inherits: trie // changed default to trie
{code}
JFR allocation report:
!jfr_allocation_profile.png|width=570!
!image-2024-12-23-09-53-27-226.png|width=570!
was:
Currently for memtable_allocation_type: offheap_objects we allocate memory to
calculate dataSize() during a write to a memtable because we use default
implementation of dataSize() which retrieves an object using value() method.
To avoid such memory allocation we can calculate data size directly for
NativeCell and NativeClustering
test schema details:
{code}
CREATE TABLE memory_tests.test_table_text_0 (part_key text,clust_key
text,value_0 text,value_1 text,value_2 text,value_3 text,value_4 text,value_5
text,value_6 text,value_7 text,value_8 text,value_9 text,PRIMARY
KEY(part_key,clust_key))
{code}
write query:
{code}
INSERT INTO memory_tests.test_table_text_0
(part_key,clust_key,value_0,value_1,value_2,value_3,value_4,value_5,value_6,value_7,value_8,value_9)
VALUES (?,?,?,?,?,?,?,?,?,?,?,?)
{code}
changed settings:
{code}
memtable_allocation_type: offheap_objects
memtable:
configurations:
skiplist:
class_name: SkipListMemtable
trie:
class_name: TrieMemtable
default:
inherits: trie // changed default to trie
{code}
!jfr_allocation_profile.png!
!image-2024-12-23-09-53-27-226.png!
> Avoid memory allocation in NativeCell.valueSize() and
> NativeClustering.dataSize()
> ---------------------------------------------------------------------------------
>
> Key: CASSANDRA-20162
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20162
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Local/Memtable
> Reporter: Dmitry Konstantinov
> Assignee: Dmitry Konstantinov
> Priority: Normal
> Fix For: 5.0.x, 5.1
>
> Attachments: image-2024-12-23-09-53-27-226.png,
> jfr_allocation_profile.png
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Currently for memtable_allocation_type: offheap_objects we allocate memory to
> calculate dataSize() during a write to a memtable because we use default
> implementation of dataSize() which retrieves an object using value() method.
> To avoid such memory allocation we can calculate data size directly for
> NativeCell and NativeClustering
> test schema details:
> {code:java}
> CREATE TABLE memory_tests.test_table_text_0 (part_key text,clust_key
> text,value_0 text,value_1 text,value_2 text,value_3 text,value_4 text,value_5
> text,value_6 text,value_7 text,value_8 text,value_9 text,PRIMARY
> KEY(part_key,clust_key))
> {code}
> write query:
> {code:java}
> INSERT INTO memory_tests.test_table_text_0
> (part_key,clust_key,value_0,value_1,value_2,value_3,value_4,value_5,value_6,value_7,value_8,value_9)
> VALUES (?,?,?,?,?,?,?,?,?,?,?,?)
> {code}
> changed settings:
> {code:java}
> memtable_allocation_type: offheap_objects
> memtable:
> configurations:
> skiplist:
> class_name: SkipListMemtable
> trie:
> class_name: TrieMemtable
> default:
> inherits: trie // changed default to trie
> {code}
> JFR allocation report:
> !jfr_allocation_profile.png|width=570!
> !image-2024-12-23-09-53-27-226.png|width=570!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
