We have around 100 snapshots for a given backupID. I am running below POST to restore each snapshot one by one as concurrency is not supported. This is the case where my entire system crashed and I am recovering to known state.
POST /_snapshot/my_backup/snapshot_1/_restore
I have around 2 TB of data to restore... In this restore stage, I don't accept any requests (either ingest or search) and my entire goal is to get complete restore as fast as possible...
I have set max_restore_bytes_per_sec=0 and performance improved a lot. Want to know any more tunings as such.
Why do you need to restore so many snapshots to bring your system back into a known state? Although you can't restore more than one snapshot at once, each snapshot can contain multiple indices and within each single restore those indices will be recovered concurrently. Therefore one possible optimisation for the future is to avoid the need to restore multiple snapshots like this.
Hi David, Thanks for response. In my case, many customers share a single setup. Each Customer creates their own indexes. On average we have 100 customers each having 20 indexes. We are trying keep a separate snapshot for each customer.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.