mirrors/scylladb
1
0
Fork 0
mirror of https://github.com/scylladb/scylladb.git synced 2026-05-12 19:02:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

repair: Improve estimated_partitions to reduce memory usage

Browse Source 
Currently, we use the sum of the estimated_partitions from each
participant node as the estimated_partitions for sstable produced by
repair. This way, the estimated_partitions is the biggest possible
number of partitions repair would write.

Since repair will write only the difference between repair participant
nodes, using the biggest possible estimation will overestimate the
partitions written by repair, most of the time.

The problem is that overestimated partitions makes the bloom filter
consume more memory. It is observed that it causes OOM in the field.

This patch changes the estimation to use a fraction of the average
partitions per node instead of sum. It is still not a perfect estimation
but it already improves memory usage significantly.

Fixes #18140

Closes scylladb/scylladb#18141
This commit is contained in:
Asias He
2024-04-02 09:21:05 +08:00
committed by Botond Dénes
parent e431e7dc16
commit 642f9a1966
1 changed files with 20 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
repair/row_level.cc
View File

@@ -3035,6 +3035,26 @@ public:
});
}).get();
if (!master.all_nodes().empty()) {
// Use the average number of partitions, instead of the sum
// of the partitions, as the estimated partitions in a
// given range. The bigger the estimated partitions, the
// more memory bloom filter for the sstable would consume.
_estimated_partitions /= master.all_nodes().size();
// In addition, estimate the difference between nodes is
// less than 10% for regular repair. Underestimation will
// not be a big problem since those sstables produced by
// repair will go through off-strategy later anyway. The
// worst case is that we have a worse false positive ratio
// than expected temporarily when the sstable is still in
// maintenance set.
//
// To save memory and have less different conditions, we
// use the 10% estimation for RBNO repair as well.
_estimated_partitions /= 10;
}
parallel_for_each(master.all_nodes(), [&, this] (repair_node_state& ns) {
const auto& node = ns.node;
rlogger.trace("Get repair_set_estimated_partitions for node={}, estimated_partitions={}", node, _estimated_partitions);