[enhancement](profile) Store profile on disk so that we can hold more profile in memory #33690
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…is into refactor-profile-be-pr
…r-profile-be-pr
|
Thank you for your contribution to Apache Doris. Since 2024-03-18, the Document has been moved to doris-website. |
zhiqiang-hhhh
force-pushed
the
profile-persist
branch
from
0f9e461 to
b4822f9
Compare
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
…doris into refactor-profile-fe
…r-profile-be-pr
…is into refactor-profile-fe
yiguolei
reviewed
Jul 24, 2024
| private static final Logger LOG = LogManager.getLogger(ProfileManager.class); | ||
| private static volatile ProfileManager INSTANCE = null; | ||
| private static final String PROFILE_STORAGE_PATH = Config.audit_log_dir + File.separator + "profile"; |
There was a problem hiding this comment.
这里默认值可以是这个，但是应该有单独的配置项，来单独设置profile 的存储路径，如果设置了，那么就不应该用这个默认的路径了
yiguolei
reviewed
Jul 24, 2024
fe/fe-core/src/main/java/org/apache/doris/common/util/ProfileManager.java
Show resolved
Hide resolved
yiguolei
reviewed
Jul 24, 2024
| @@ -2694,6 +2694,10 @@ public class Config extends ConfigBase { | |||
| @ConfField(mutable = true) | |||
| public static boolean enable_cooldown_replica_affinity = true; | |||
|
|
|||
| // The max number of profiles that can be stored to storage. | |||
| @ConfField | |||
| public static int max_spilled_profile_num = 500; | |||
There was a problem hiding this comment.
不能用个数限制，应该用磁盘空间大小来限制。 因为个数用户说不清楚。
There was a problem hiding this comment.
个数限制 + 磁盘大小可能都得需要，个数限制防止产生大量小文件。我们后台有很多 select @@version_comment limit 1 类似这样的查询。
yiguolei
reviewed
Jul 24, 2024
| readLock.lock(); | ||
| try { | ||
| ExecutionProfile execProfile = queryIdToExecutionProfiles.get(queryId); | ||
| if (execProfile == null) { |
There was a problem hiding this comment.
一个profile 对应多个execution profile，这里不要从profilemanager 里读取了，直接profile
对象里就包含多个execution profile，直接都设置了吧
There was a problem hiding this comment.
这里拿不到 Profile 对象，入参是 TUniqueId，只能对应 ExecutionProfile
yiguolei
reviewed
Jul 24, 2024
| @@ -167,18 +167,13 @@ public void unregisterQuery(TUniqueId queryId) { | |||
| if (LOG.isDebugEnabled()) { | |||
| LOG.debug("Deregister query id {}", DebugUtil.printId(queryId)); | |||
| } | |||
| ExecutionProfile executionProfile = ProfileManager.getInstance().getExecutionProfile(queryId); | |||
| if (executionProfile != null) { | |||
| executionProfile.setQueryFinishTime(System.currentTimeMillis()); | |||
There was a problem hiding this comment.
为什么要动这里，这里只跟cood 有关，跟profile 本身不相关。
There was a problem hiding this comment.
逻辑没变，只是把对 ExecutionProfile 的操作都通过 ProfileManager 去做了
|
run buildall |
TPC-H: Total hot run time: 40065 ms |
TPC-DS: Total hot run time: 172883 ms |
ClickBench: Total hot run time: 31.24 s |
yiguolei
approved these changes
Jul 25, 2024
|
PR approved by at least one committer and no changes requested. |
github-actions
bot
added
the
approved
|
PR approved by anyone and no changes requested. |
Gabriel39
approved these changes
Jul 25, 2024
hello-stephen
pushed a commit
that referenced
this pull request
Jul 26, 2024
dataroaring
pushed a commit
that referenced
this pull request
Aug 11, 2024
… profile in memory (#33690) Step 3 of #33744 Store profile on disk of FE, in the same path with audit log. 1. ProfileManager will have a daemon thread that checks whether a profile should be stored on disk 2. Which profile can be stored? a. query finished. b. collection of profile has finished or query itself has finished for a long time, like 5 seconds 3. Profile structure on disk: ``` * Integer: n(size of summary profile) * String: json of summary profile * Integer: m(size of compressed execution profile) * String: compressed binary of execution profile ``` 4. IO thread of ProfileManager will also remove garbage from disk. 5. Once a profile is stored to disk, its detail content will be release from memory, so that we can hold more profiles in memory, further access to profile will read from disk directly. 6. Basic but necessary UT for profile serialization. 7. Refine constructor of ExecutionProfile. Further work: 1. Abstract class ProfileReader/ProfileWriter, they define common interface for profile io, and we can implement ProfileDiskReader/ProfileDiskWriter and also ProfileS3Reader/ProfileS3Writer to support profile io on object storage. 2. Finer granularity for profile creation. Currently for ddl like `create/drop table`, profile is also created, this is unnecessary. 3. More regression test. 4. More reasonable default value of max_profile_on_disk. 5. system table for profile storage, so that we can figure out how much storage is costed for profile --------- Co-authored-by: yiguolei <[email protected]> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
dataroaring
pushed a commit
that referenced
this pull request
Aug 12, 2024
dataroaring
pushed a commit
that referenced
this pull request
Aug 16, 2024
… profile in memory (#33690) Step 3 of #33744 Store profile on disk of FE, in the same path with audit log. 1. ProfileManager will have a daemon thread that checks whether a profile should be stored on disk 2. Which profile can be stored? a. query finished. b. collection of profile has finished or query itself has finished for a long time, like 5 seconds 3. Profile structure on disk: ``` * Integer: n(size of summary profile) * String: json of summary profile * Integer: m(size of compressed execution profile) * String: compressed binary of execution profile ``` 4. IO thread of ProfileManager will also remove garbage from disk. 5. Once a profile is stored to disk, its detail content will be release from memory, so that we can hold more profiles in memory, further access to profile will read from disk directly. 6. Basic but necessary UT for profile serialization. 7. Refine constructor of ExecutionProfile. Further work: 1. Abstract class ProfileReader/ProfileWriter, they define common interface for profile io, and we can implement ProfileDiskReader/ProfileDiskWriter and also ProfileS3Reader/ProfileS3Writer to support profile io on object storage. 2. Finer granularity for profile creation. Currently for ddl like `create/drop table`, profile is also created, this is unnecessary. 3. More regression test. 4. More reasonable default value of max_profile_on_disk. 5. system table for profile storage, so that we can figure out how much storage is costed for profile --------- Co-authored-by: yiguolei <[email protected]> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
dataroaring
pushed a commit
that referenced
this pull request
Aug 16, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Step 3 of #33744
Store profile on disk of FE, in the same path with audit log.
a. query finished.
b. collection of profile has finished or query itself has finished for a long time, like 5 seconds
Further work:
create/drop table, profile is also created, this is unnecessary.