Add rack-aware load balancing policy#73
Merged
jul-stas merged 5 commits intoscylladb:masterfrom Jan 18, 2023
Merged
Conversation
The out was not helpful for understanding the failure of a test. Printing the query plans will help.
This argument was not used anywhere, it worked by chance since it had the same value everywhere.
The Datstax C++ driver has default ONE, the Scylla fork has LOCAL_ONE.
We need to prefer local rack as there are higher network costs when communicating with nodes in remote rack. This policy prefers nodes from the local rack, then local datacenter and then other nodes. The new RackAwarePolicy is similar to DCAwarePolicy, but does not have the deprecated options. TokenAwarePolicy and other code needed to be modified so that the local rack is propagated. The TokenAware policy was changed to prefer replicas in remote rack / remote DC before trying non-replica nodes. It does not make much sense to not try the replicas and trying the replicas simplifies the code as now we have three levels local/remote/remote2. This change might not be backwards-compatible, we don't know what exactly this project guarantees in terms of backwards compatibility. Co-Authored-By: Peter Navrátil <peter.navratil@kiwi.com>
martin-sucha
force-pushed
the
ms/rack-aware-upstream
branch
from
5c444f9 to
9691ec0
Compare
Author
|
See also apache#536 |
avelanarius
reviewed
Jan 4, 2023
Comment on lines
+42
to
+62
| if (local_dc_.empty()) { // Only override if no local DC was specified. | ||
| local_dc_ = local_dc; | ||
| } | ||
|
|
||
| if (local_dc_.empty() && connected_host && !connected_host->dc().empty()) { | ||
| LOG_INFO("Using '%s' for the local data center " | ||
| "(if this is incorrect, please provide the correct data center)", | ||
| connected_host->dc().c_str()); | ||
| local_dc_ = connected_host->dc(); | ||
| } | ||
|
|
||
| if (local_rack_.empty()) { // Only override if no local rack was specified. | ||
| local_rack_ = local_rack; | ||
| } | ||
|
|
||
| if (local_rack_.empty() && connected_host && !connected_host->rack().empty()) { | ||
| LOG_INFO("Using '%s' for the local rack " | ||
| "(if this is incorrect, please provide the correct rack)", | ||
| connected_host->rack().c_str()); | ||
| local_rack_ = connected_host->rack(); | ||
| } |
There was a problem hiding this comment.
Minor issue: when only one of local_dc or local_rack is empty, the empty one is populated from connected_host (which corresponds to a control connection) while the non-empty one is populated from a user configuration, which could result in nonsensical DC+rack combination.
avelanarius
approved these changes
Jan 4, 2023
avelanarius
left a comment
avelanarius
left a comment
There was a problem hiding this comment.
Apart from the one minor problem, the policy looks correct to me.
4 tasks
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 free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
We need to prefer local rack as there are higher network costs
when communicating with nodes in remote rack.
This policy prefers nodes from the local rack, then local datacenter
and then other nodes.
The new RackAwarePolicy is similar to DCAwarePolicy,
but does not have the deprecated options.
TokenAwarePolicy and other code needed to be modified
so that the local rack is propagated.
The TokenAware policy was changed to prefer replicas in remote
rack / remote DC before trying non-replica nodes.
It does not make much sense to not try the replicas and
trying the replicas simplifies the code as now we have three
levels local/remote/remote2.
This change might not be backwards-compatible,
we don't know what exactly this project guarantees in terms of
backwards compatibility.
Co-Authored-By: Peter Navrátil peter.navratil@kiwi.com