Thank you all, that’s very helpful. It looks like there’s something else that’s causing my cluster to not load my savepoints, so I’ve submitted a separate query for that.
Many thanks, John Sent from my iPhone On 6 Jul 2022, at 21:24, Alexander Fedulov <alexan...@ververica.com> wrote: Hi John, use $ bin/flink run -s s3://my_bucket/path/to/savepoints/<savepoint id> (no trailing slash, including schema). where <savepoint id> should contain a valid _metadata file. You should see logs like this: INFO o.a.f.r.c.CheckpointCoordinator [] - Starting job foobar from savepoint s3://my_bucket/path/to/savepoints/<savepoint id> () INFO o.a.f.r.c.CheckpointCoordinator [] org.apache.flink.runtime.checkpoint.CheckpointCoordinator [] - Restoring job foobar from Savepoint 1 @ 0 for foobar located at s3://my_bucket/path/to/savepoints/<savepoint id>. The indication of the correct restore should be the absence of exceptions. You might see messages like this one for operators that did not have any state in the savepoint: INFO o.a.f.r.c.CheckpointCoordinator [] - Skipping empty savepoint state for operator a0f11f7a2c416beb6b7aed14be0d63ca. Best, Alexander Fedulov On Wed, Jul 6, 2022 at 9:50 PM John Tipper <john_tip...@hotmail.com<mailto:john_tip...@hotmail.com>> wrote: Hi all, The docs on restoring a job from a savepoint (https://nightlies.apache.org/flink/flink-docs-master/docs/ops/state/savepoints/#resuming-from-savepoints) state that the syntax is: $ bin/flink run -s :savepointPath [:runArgs] and where "you may give a path to either the savepoint’s directory or the _metadata file." If I am using S3 as my store of state: state.savepoints.dir: s3://my_bucket/path/to/savepoints and an example savepoint is at: s3://my_bucket/path/to/savepoints/<savepoint id>/_metadata then what am I supposed to supply to the flink run command? Is it: 1. The full path including filesystem: s3://my_bucket/path/to/savepoints/<savepoint id>/_metadata or s3://my_bucket/path/to/savepoints/<savepoint id> 2. or the full path: my_bucket/path/to/savepoints/<savepoint id>/_metadata or my_bucket/path/to/savepoints/<savepoint id> 3. or the path relative to the savepoints directory: <savepoint id>/_metadata or <savepoint id> If I supply a directory, do I need to specify a trailing slash? Also, is there anything that I will see in the logs that will indicate that the restore from a savepoint has been successful? Many thanks, John