Does repair only read rows from disk?

ScyllaDB
,
1

Installation details
#ScyllaDB version: all
#Cluster size:
os (RHEL/CentOS/Ubuntu/AWS AMI):
Recently I read repair code about row_level_epair, and I see it has a function called read_rows_from_disk, then, I am wondering will it read data from memtable? Or only sstable ?

Here is an example, when there is a five-node cluster with 3 replicas, consistency is quorum, and it is doing repair, considering at the same time a write is settled to disk(sstable) on node-1, this write is still on memtable of the other two nodes. So, in this situation, how repair reads data ?

If repair only reads data from sstable, will repair thinks that write is missed on the other two nodes?

2

Correct cluster config: six-node cluster with 5 replicas

3

Repair reads both from disk and memtables. Most data is expected to come from the disk, so repair code refers to data as coming from disk.

4

Thanks a lot for your reply. And I see data is read to a cache with a limit value, then do a hash for it. So, if an item is stored in several sstable, how it compare the diff?

5

Data from different SSTables and Memtables are merged into a single unified stream, this is what is stored in the repair buffers. This buffer is then used to compare against the unified stream on other replicas.

1 Like