[Fix](merge-on-write) Fix FE may use the staled response to wrongly commit txn #39018
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
|
Thank you for your contribution to Apache Doris. Since 2024-03-18, the Document has been moved to doris-website. |
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
bobhan1
force-pushed
the
fix-commit-txn-with-wrong-calc-result
branch
from
c60d7c4
to
a6110ad
Compare
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
PR approved by at least one committer and no changes requested. |
github-actions
bot
added
approved
|
PR approved by anyone and no changes requested. |
Yukang-Lian
approved these changes
Aug 7, 2024
zhannngchen
requested changes
Aug 7, 2024
github-actions
bot
removed
the
approved
|
clang-tidy review says "All clean, LGTM! 👍" |
2 similar comments
|
clang-tidy review says "All clean, LGTM! 👍" |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
run buildall |
|
clang-tidy review says "All clean, LGTM! 👍" |
TPC-H: Total hot run time: 39793 ms |
…tinel mark in _ms_base_compaction_cnt
zhannngchen
reviewed
Aug 8, 2024
...ion-test/suites/fault_injection_p0/cloud/test_cloud_mow_stale_resp_load_load_conflict.groovy
Outdated
Show resolved
Hide resolved
|
run buildall |
github-actions
bot
added
the
approved
|
PR approved by at least one committer and no changes requested. |
TPC-H: Total hot run time: 39598 ms |
TPC-DS: Total hot run time: 205221 ms |
ClickBench: Total hot run time: 30.79 s |
github-actions
bot
removed
the
approved
|
run buildall |
|
PR approved by at least one committer and no changes requested. |
github-actions
bot
added
the
approved
|
run performance |
TPC-H: Total hot run time: 39424 ms |
TPC-DS: Total hot run time: 200924 ms |
ClickBench: Total hot run time: 31.01 s |
dataroaring
pushed a commit
that referenced
this pull request
Aug 10, 2024
…ommit txn (#39018) ## Problem consider the following scenarios for merge-on-write table in cloud mode ### Scenario 1: Load-Load Conflict 1. load txn1 tries to commit version n and gets the delete bitmap update lock 2. load txn1 begins to calculate delete bitmap on BEs, this is a heavy calculating process and lasts long 3. load txn2 tries to commit version n and gets the delete bitmap update lock because load txn1's delete bitmap update lock has expired 4. load txn1's delete bitmap update lock expires and load txn2 get the delete bitmap update lock 5. load txn2 commits successfully with version n and release the delete bitmap update lock 6. load txn1 fails to commit due to timeout of the calculation of delete bitmap 7. load txn1 retries the commit process with version n+1, gets the bitmap update lock and sends delete bitmap calculation task to BEs 8. BE fails to register this new calculation task because there is a task with the same signatrure(txn_id) running in the task_worker_pool 9. BE finishes the calculation of delete bitmap and report success status to FE 10. load txn1 commits successfully with n+1 Finally, load txn1 failed to calculate delete bitmap for version n from load txn2 ### Scenario 2: Load-Compaction Conflict 1. load txn tries to commit and gets the delete bitmap update lock 2. load txn collects rowset_ids and submit a delete bitmap calculation task to the threadpool for the diff rowsets. But the theadpool is full, so the task is queued in the threadpool. 3. load txn's delete bitmap update lock expired and a compaction job on the same tablet finished successfully. 4. load txn fails to commit due to timeout of the calculation of delete bitmap 5. load txn retries the commit process, gets the bitmap update lock and sends delete bitmap calculation task to BEs 6. BE fails to register this new calculation task because there is a task with the same signatrure(txn_id) running in the task_worker_pool 7. BE finishes the calculation of delete bitmap and report success status to FE 8. load txn1 commits successfully Finally, load txn failed to calculate delete bitmap for the compaction produced by compaction ## Solution The root cause of the above failures is that when the commit process is retried many times, FE may use the previous stale success response from BEs and commit txns. One solution for that problem is that FE attaches an unique id within the delete bitmap calculation task sent to BE and BE takes it in the response for FE to check if the response is for the current latest task. However, if the delete bitmap calculation always consumes more time than the timeout of the delete bitmap calculation, FE will retry the commit process infinitely which causes live lock. This PR let the BE's response take the compaction stats(to avoid load-compaction conflict) and versions(to avoid load-load conflict) from the task request and let the FE compares it with the current task's to know that if there is any compaction or load finished during the time periods since the current load get the delete bitmap lock due to lock expiration. If so, the current txn should retry or abort. If not, the current txn can commit successfully.
wyxxxcat
pushed a commit
to wyxxxcat/doris
that referenced
this pull request
Aug 14, 2024
…ommit txn (apache#39018) ## Problem consider the following scenarios for merge-on-write table in cloud mode ### Scenario 1: Load-Load Conflict 1. load txn1 tries to commit version n and gets the delete bitmap update lock 2. load txn1 begins to calculate delete bitmap on BEs, this is a heavy calculating process and lasts long 3. load txn2 tries to commit version n and gets the delete bitmap update lock because load txn1's delete bitmap update lock has expired 4. load txn1's delete bitmap update lock expires and load txn2 get the delete bitmap update lock 5. load txn2 commits successfully with version n and release the delete bitmap update lock 6. load txn1 fails to commit due to timeout of the calculation of delete bitmap 7. load txn1 retries the commit process with version n+1, gets the bitmap update lock and sends delete bitmap calculation task to BEs 8. BE fails to register this new calculation task because there is a task with the same signatrure(txn_id) running in the task_worker_pool 9. BE finishes the calculation of delete bitmap and report success status to FE 10. load txn1 commits successfully with n+1 Finally, load txn1 failed to calculate delete bitmap for version n from load txn2 ### Scenario 2: Load-Compaction Conflict 1. load txn tries to commit and gets the delete bitmap update lock 2. load txn collects rowset_ids and submit a delete bitmap calculation task to the threadpool for the diff rowsets. But the theadpool is full, so the task is queued in the threadpool. 3. load txn's delete bitmap update lock expired and a compaction job on the same tablet finished successfully. 4. load txn fails to commit due to timeout of the calculation of delete bitmap 5. load txn retries the commit process, gets the bitmap update lock and sends delete bitmap calculation task to BEs 6. BE fails to register this new calculation task because there is a task with the same signatrure(txn_id) running in the task_worker_pool 7. BE finishes the calculation of delete bitmap and report success status to FE 8. load txn1 commits successfully Finally, load txn failed to calculate delete bitmap for the compaction produced by compaction ## Solution The root cause of the above failures is that when the commit process is retried many times, FE may use the previous stale success response from BEs and commit txns. One solution for that problem is that FE attaches an unique id within the delete bitmap calculation task sent to BE and BE takes it in the response for FE to check if the response is for the current latest task. However, if the delete bitmap calculation always consumes more time than the timeout of the delete bitmap calculation, FE will retry the commit process infinitely which causes live lock. This PR let the BE's response take the compaction stats(to avoid load-compaction conflict) and versions(to avoid load-load conflict) from the task request and let the FE compares it with the current task's to know that if there is any compaction or load finished during the time periods since the current load get the delete bitmap lock due to lock expiration. If so, the current txn should retry or abort. If not, the current txn can commit successfully.
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.
Problem
consider the following scenarios for merge-on-write table in cloud mode
Scenario 1: Load-Load Conflict
Finally, load txn1 failed to calculate delete bitmap for version n from load txn2
Scenario 2: Load-Compaction Conflict
Finally, load txn failed to calculate delete bitmap for the compaction produced by compaction
Solution
The root cause of the above failures is that when the commit process is retried many times, FE may use the previous stale success response from BEs and commit txns. One solution for that problem is that FE attaches an unique id within the delete bitmap calculation task sent to BE and BE takes it in the response for FE to check if the response is for the current latest task. However, since
calculate_delete_bitmap_task_timeout_secondscan not change adaptively based on actual computation time currently, if the delete bitmap calculation always consumes more time than the timeout of the delete bitmap calculation, FE will retry the commit process infinitely which causes live lock.This PR let the BE's response take the compaction stats(to avoid load-compaction conflict) and versions(to avoid load-load conflict) from the task request and let the FE compares it with the current task's to know that if there is any compaction or load finished during the time periods since the current load get the delete bitmap lock due to lock expiration. If so, the current txn should retry or abort. If not, the current txn can commit successfully.