-
Notifications
You must be signed in to change notification settings - Fork 1.5k
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
[BREAKING] perf(Transactions): Run transactions concurrently #7694
Merged
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
manishrjain
changed the title
opt(Transactions): Run transactions concurrently
perf(Transactions): Run transactions concurrently
Apr 9, 2021
manishrjain
changed the title
perf(Transactions): Run transactions concurrently
[BREAKING] perf(Transactions): Run transactions concurrently
Apr 10, 2021
manishrjain
added a commit
that referenced
this pull request
May 5, 2021
In this PR, we use concurrent mutations system, introduced in #7694 , to generate Skiplists directly from a transaction, even before it is committed. When a transaction does get committed, we replace the key timestamps with the commit timestamp, in-place within the Skiplist. And in case of multiple transactions being committed, merge them together into one bigger Skiplist during the serial commit. This Skiplist is then handed over to Badger, bypassing its value log and WAL system. Instead, when Badger persists the skiplist into L0 tables, Dgraph gets a callback, which it uses to decide when it is safe to take a Raft snapshot / checkpoint. So, we no longer need Badger's WAL. Furthermore, live loader is also optimized to better deal with the conflicting transactions, and retrying them as frequently as possible, instead of only towards the end. The updated output from live loader shows we're able to deal with stragglers much better, with them causing almost no delays with this PR. There's also a bug fix in this PR. The Raft checkpointing code was not getting run as often as it should. This PR fixes that. ``` Master: [18:48:34-0700] Elapsed: 08m45s Txns: 21222 N-Quads: 21,221,870 N-Quads/s: 9,033 Aborts: 0 Number of TXs run : 21240 Number of N-Quads processed : 21239870 Time spent : 8m49.039284385s N-Quads processed per second : 40150 This PR: Number of TXs run : 21240 Number of N-Quads processed : 21239870 Time spent : 6m56.641399434s N-Quads processed per second : 51057 ``` Notable Changes: * Have each mutation create its own skiplist. Merge them during commit. * Use skiplist Builder * optimize mutations as well * Optimize live loader to better deal with conflicted requests. * Make a callback delete txns from Oracle. * Bug fix: Calculate raft checkpoint frequently. * Clarify snapshot rules. * Incremental rollups use skiplists too. * Stop rollups during drop operations. Co-authored-by: Ahsan Barkati <[email protected]>
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.
This PR introduced concurrent mutation processing. It does so in this way:
This sequence improves the 75-th %-ile latency of mutations from 30ms down to 1ms. The 21M live loader dataset gets loaded on my workstation in 8 mins.
Given these improvements, I've removed ludicrous mode, which was buggy. This PR also contains a bug fix for how we were handling filterStringFunction.
This change is