How to efficiently dump and restore a database snapshot?

[Benedikt Waldvogel]

Hello,

we’re using H2 in some integration tests.

We have a rather long-running setup phase that populates the database with a basic set of data that can be used for a large variety of tests.

To speed-up the integration tests, we run the setup only once, create a snapshot and restore it before each test to keep the tests fully isolated.

We basically run "SCRIPT TO '/path/to/dump.zip'" to create the snapshot and "RUNSCRIPT FROM '/path/to/dump.zip'" to restore it.

Profiling indicates that restoring the snapshot can take a significant fraction of test time, especially if the test itself is small and runs quickly.

I’m wondering if there is a more efficient way to dump and restore the H2 database?

Is it possible to access the underlying MVStore and dump it to a file/directory and later replace the MVStore with the contents of the dump?

Any hint the points me in the right direction is appreciated.

Best regards

Benedikt Waldvogel

[Noel Grandin]

just make a file copy of the underlying database and copy it back for each test?

[Benedikt Waldvogel]

Hi Noel,

On Thursday, March 7, 2019 at 7:15:18 AM UTC+1, Noel Grandin wrote:

just make a file copy of the underlying database and copy it back for each test?

We currently have the database in-memory but I’ll give it a try to switch to on-disk, thanks.

However, I’ll need to measure the performance impact of having the DB on-disk instead of in-memory.

Do you have an idea how to achieve it when the database is in-memory?

Thanks

Benedikt

[Noel Grandin]

>
> Do you have an idea how to achieve it when the database is in-memory?
>

Our byte storage level backends all emulate a kind of filesystem abstraction, so in theory you could copy in-memory
files around too.

See the org.h2.store.fs.Filepath class

You'd need to do some experimentation, and you might wander into a bug, because I don't think anyone has ever done that
before :-)