-
Notifications
You must be signed in to change notification settings - Fork 445
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Offers new ways to compute bulk import load plans. #4933
Open
keith-turner
wants to merge
6
commits into
apache:3.1
Choose a base branch
from
keith-turner:bulk-load-improvement-31
base: 3.1
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
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
Two new ways of computing bulk import load plans are offered in these change. First the RFile API was modified to support computing a LoadPlan as the RFile is written. Second a new LoadPlan.compute() method was added that creates a LoadPlan from an existing RFile. In addition to these changes methods were added to LoadPlan that support serializing and deserializing load plans to/from json. All of these changes together support the use case of computing load plans in a distributed manner. For example, with a bulk import directory with N files the following use case is now supported. 1. For eack file a task is spun up on a remote server that calls the new LoadPlan.compute() API to determine what tablets the file overlaps. Then the new LoadPlan.toJson() method is called to serialize the load plan and send it to a central place. 2. All the load plans from the remote servers are deserialized calling the new LoadPlan.fromJson() method and merged into a single load plan that is used to do the bulk import. Another use case these new APIs could support is running this new code in the map reduce job that generates bulk import data. 1. In each reducer as it writes to an rfile it could also be building a LoadPlan. A load plan can be obtained from the Rfile after closing it and serialized using LoadPlan.toJson() and the result saved to a file. So after the map reduce job completes each rfile would have corresponding file with a load plan for that file. 2. Another process that runs after the map reduce job can load all the load plans from files and merge them using the new LoadPlan.fromJson() method. Then the merged LoadPlan can be used to do the bulk import. BulkNewIT.testComputeLoadPlan() simulates this map reduce use case by going through the steps in code that a map reduce job would. This tests the new APIs and shows what using it would look like. Both of these use cases avoid doing the analysis of files on a single machine doing the bulk import. Bulk import V1 had this functionality and would ask random tservers to do the file analysis. This could cause unexpected load on those tservers. Bulk V1 would interleave analyzing files and adding them to tablets. This could lead to odd situations where files are partially imported to some tablets and analysis fails, leaving the file partially imported. Bulk v2 does all analysis before any files are added to tablets, however it lacks this distributed analysis capability. These changes provide the building blocks to do the distributed analysis that bulk v1 did for bulk v2.
This PR squashed the changes from #4898 into 3.1 and resolved conflicts. |
keith-turner
commented
Sep 30, 2024
@@ -92,12 +93,17 @@ public class RFileWriter implements AutoCloseable { | |||
|
|||
private final FileSKVWriter writer; | |||
private final LRUMap<ByteSequence,Boolean> validVisibilities; | |||
|
|||
// TODO should be able to completely remove this as lower level code is already doing some things |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I will remove this todo and open an issue before this PR is merged.
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.
Two new ways of computing bulk import load plans are offered in these change. First the RFile API was modified to support computing a LoadPlan as the RFile is written. Second a new LoadPlan.compute() method was added that creates a LoadPlan from an existing RFile. In addition to these changes methods were added to LoadPlan that support serializing and deserializing load plans to/from json.
All of these changes together support the use case of computing load plans in a distributed manner. For example, with a bulk import directory with N files the following use case is now supported.
Another use case these new APIs could support is running this new code in the map reduce job that generates bulk import data.
Both of these use cases avoid doing the analysis of files on a single machine doing the bulk import. Bulk import V1 had this functionality and would ask random tservers to do the file analysis. This could cause unexpected load on those tservers. Bulk V1 would interleave analyzing files and adding them to tablets. This could lead to odd situations where files are partially imported to some tablets and analysis fails, leaving the file partially imported. Bulk v2 does all analysis before any files are added to tablets, however it lacks this distributed analysis capability. These changes provide the building blocks to do the distributed analysis that bulk v1 did for bulk v2.