Uh oh!
There was an error while loading. Please reload this page.
[improvement](tablet scheduler) Adjust tablet sched priority to help load data succ - #38528
Merged
Merged
Conversation
doris-robot
commented
Jul 30, 2024
Thank you for your contribution to Apache Doris. Since 2024-03-18, the Document has been moved to doris-website. |
yujun777
commented
Jul 30, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Jul 30, 2024
TPC-H: Total hot run time: 41523 ms |
doris-robot
commented
Jul 30, 2024
TPC-DS: Total hot run time: 169211 ms |
yujun777
commented
Jul 30, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Jul 30, 2024
ClickBench: Total hot run time: 30.07 s |
doris-robot
commented
Jul 30, 2024
TPC-H: Total hot run time: 41505 ms |
doris-robot
commented
Jul 30, 2024
TPC-DS: Total hot run time: 169586 ms |
doris-robot
commented
Jul 30, 2024
ClickBench: Total hot run time: 29.84 s |
yujun777force-pushed
the
impr-tablet-sched-2
branch
from
July 30, 2024 10:28
acb18c8 to
14717f6Compareyujun777
commented
Jul 30, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Jul 30, 2024
TPC-H: Total hot run time: 41850 ms |
doris-robot
commented
Jul 30, 2024
TPC-DS: Total hot run time: 169797 ms |
doris-robot
commented
Jul 30, 2024
ClickBench: Total hot run time: 30.29 s |
yujun777
commented
Jul 30, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Jul 30, 2024
TPC-H: Total hot run time: 43523 ms |
doris-robot
commented
Jul 30, 2024
TPC-DS: Total hot run time: 169172 ms |
doris-robot
commented
Jul 30, 2024
ClickBench: Total hot run time: 30.08 s |
yujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
yujun777force-pushed
the
impr-tablet-sched-2
branch
from
July 31, 2024 02:33
25580c9 to
499046cCompareyujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
3 similar comments
yujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
yujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
yujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
yujun777
commented
Jul 31, 2024
ContributorAuthor
run feut |
yujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
yujun777force-pushed
the
impr-tablet-sched-2
branch
from
July 31, 2024 09:03
b1cbeea to
84bf2b8Compareyujun777
commented
Jul 31, 2024
ContributorAuthor
run buildall |
yujun777force-pushed
the
impr-tablet-sched-2
branch
from
August 1, 2024 01:47
84bf2b8 to
b2ce3c2Compareyujun777
commented
Aug 1, 2024
ContributorAuthor
run buildall |
yujun777
commented
Aug 1, 2024
ContributorAuthor
run performance |
yujun777
commented
Aug 1, 2024
ContributorAuthor
run buildall |
yujun777
commented
Aug 1, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Aug 1, 2024
TPC-H: Total hot run time: 41731 ms |
doris-robot
commented
Aug 1, 2024
TPC-DS: Total hot run time: 170219 ms |
doris-robot
commented
Aug 1, 2024
ClickBench: Total hot run time: 29.65 s |
Contributor
PR approved by at least one committer and no changes requested. |
dataroaring pushed a commit
that referenced
this pull request
Aug 11, 2024
…load data succ (#38528) improve tablet repair sched: 1. if a tablet had version incomplete replicas, then first fix the version incomplete replicas, after that if tablet still no enough replicas, then add new replicas. This improvement will also fix the bug: for 3 replica on 3 backend, if one replica A 's backend is dead, one replica B miss versions, then the tablet's status is REPLICA_MISSING. Since no new backend to locate a new replica, the sched will always fail, also the missing versions B will not be repair. This PR will try to fix replica B firstly, only after that it then try to add a new replica. 2. when load data fail, then repair this tablet imm; 3. increase those tablets' sched priority: a) recently write failed; b) had version incomplete replicas; c) mow; 4. fix colocate table health status, if colocate tablet's replica are not alive, its status should be unrecoverable. But adjust priority is still not enough due to the sched pending queue limit size 2000. TabletChecker will put 2000 sched tasks into the sched pending queue, if the queue is full, even the highest priority sched task couldn't put into the queue. It need to wait until the sched pending queue is not full later.
dataroaring pushed a commit
that referenced
this pull request
Aug 16, 2024
…load data succ (#38528) improve tablet repair sched: 1. if a tablet had version incomplete replicas, then first fix the version incomplete replicas, after that if tablet still no enough replicas, then add new replicas. This improvement will also fix the bug: for 3 replica on 3 backend, if one replica A 's backend is dead, one replica B miss versions, then the tablet's status is REPLICA_MISSING. Since no new backend to locate a new replica, the sched will always fail, also the missing versions B will not be repair. This PR will try to fix replica B firstly, only after that it then try to add a new replica. 2. when load data fail, then repair this tablet imm; 3. increase those tablets' sched priority: a) recently write failed; b) had version incomplete replicas; c) mow; 4. fix colocate table health status, if colocate tablet's replica are not alive, its status should be unrecoverable. But adjust priority is still not enough due to the sched pending queue limit size 2000. TabletChecker will put 2000 sched tasks into the sched pending queue, if the queue is full, even the highest priority sched task couldn't put into the queue. It need to wait until the sched pending queue is not full later.
HappenLee pushed a commit
to HappenLee/incubator-doris
that referenced
this pull request
Apr 24, 2026
…load data succ (apache#38528) improve tablet repair sched: 1. if a tablet had version incomplete replicas, then first fix the version incomplete replicas, after that if tablet still no enough replicas, then add new replicas. This improvement will also fix the bug: for 3 replica on 3 backend, if one replica A 's backend is dead, one replica B miss versions, then the tablet's status is REPLICA_MISSING. Since no new backend to locate a new replica, the sched will always fail, also the missing versions B will not be repair. This PR will try to fix replica B firstly, only after that it then try to add a new replica. 2. when load data fail, then repair this tablet imm; 3. increase those tablets' sched priority: a) recently write failed; b) had version incomplete replicas; c) mow; 4. fix colocate table health status, if colocate tablet's replica are not alive, its status should be unrecoverable. But adjust priority is still not enough due to the sched pending queue limit size 2000. TabletChecker will put 2000 sched tasks into the sched pending queue, if the queue is full, even the highest priority sched task couldn't put into the queue. It need to wait until the sched pending queue is not full later.
This file contains hidden or 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
Sign up for freeto 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.
improve tablet repair sched:
This improvement will also fix the bug: for 3 replica on 3 backend, if one replica A 's backend is dead, one replica B miss versions, then the tablet's status is REPLICA_MISSING. Since no new backend to locate a new replica, the sched will always fail, also the missing versions B will not be repair. This PR will try to fix replica B firstly, only after that it then try to add a new replica.
when load data fail, then repair this tablet imm;
increase those tablets' sched priority:
a) recently write failed;
b) had version incomplete replicas;
c) mow;
fix colocate table health status, if colocate tablet's replica are not alive, its status should be unrecoverable.
But adjust priority is still not enough due to the sched pending queue limit size 2000. TabletChecker will put 2000 sched tasks into the sched pending queue, if the queue is full, even the highest priority sched task couldn't put into the queue. It need to wait until the sched pending queue is not full later.