Expectation of cluster rebuild time for OpenSearch

hello. I’m wondering what folks have seen the time its taken for cluster restores or rebuilds from a snapshot with OpenSearch. I know it varies massively on size of the cluster but can some of you tell me what a restore time takes including the dataset size. Maybe also the node count? Thanks in advance. Jordan

Hey @jordanoliver,

Not exactly just on the subject, but [1] provides some helpful facts regarding snapshot size and backup/restore/reindex times. Hope it is helpful.

[1] Migration via Snapshot restore. Mapping roadblocks - #5 by jbolle

TLDR:
Test it yourself on a small subset of your data, because your network config, node load / IO constraints, etc, and repository settings will be the dominate variables.

More Words:

@jordanoliver those times were all very rough guesses. Honestly, unless someone has your same data and nodes, they can’t really tell you. And also, you can set the max restore bytes per second on the snapshot repository, so in some cases that may be the determining factor for how fast you can do a restore.
Assuming appropriately sized nodes, that aren’t busy (because they are empty, waiting for snapshot restored data), I think network bandwidth and the rate limiting on the snapshot repository will be the limiting factors. Mostly because restore from snapshot is a lot like a file copy. My issues in the linked ticket are because I had different index mappings in the cluster I was restoring into, including some plugins that weren’t available in OpenSearch.

[EDIT: added link to repository API docs]