Hey Dmitri,
> Would you be able to share any detailed performance test numbers for the
new JDBC Persistence implementation?
Please find the number of the below which i used in the above setup:
Test case | Implementation | Query | Number of
requests | Number of successful requests | Number of failed
requests | avg response time(ms) | min response time (ms)| p50
response time (ms) | max response time(ms)
---------------------|----------------|------------------|----------------------|---------------------------------|-----------------------------|---------------------|---------------------|---------------------|---------------------
Dataset creation | JDBC-PG | Create Catalog |
500 | 500 |
0 | 233 | 55 | 235 |
437
Dataset creation | JDBC-PG | Create Namespace |
65535 | 65535 |
0 | 24 | 17 | 23 |
176
Dataset creation | JDBC-PG | Create Table |
65536 | 65536 |
0 | 187 | 78 | 188 |
1131
Dataset creation | JDBC-PG | Create View |
65536 | 65536 |
0 | 74 | 116 | 117 |
1162
50/50 RW workload | JDBC-PG | Read |
14951 | 14951 |
0 | 23 | 7 | 18 |
360
50/50 RW workload | JDBC-PG | Write |
15113 | 15113 |
0 | 63 | 12 | 64 |
174
99/01 RW workload | JDBC-PG | Read |
29859 | 29859 |
0 | 18 | 7 | 14 |
401
99/01 RW workload | JDBC-PG | Write |
295 | 295 |
0 | 47 | 11 | 50 |
88
Unfortunately our mail client does not allow adding zip attachments,
please let me know if you need detailed gatling reports, happy to send
it over the slack !
Best,
Prashant Singh
On Tue, May 13, 2025 at 11:49 AM Dmitri Bourlatchkov <di...@apache.org>
wrote:
> Thanks for implementing the new JDBC persistence, Prashant!
>
> It's great to see that JDBC persistence is more stable under concurrent
> load now.
>
> Would you be able to share any detailed performance test numbers for the
> new JDBC Persistence implementation?
>
> I believe the doc linked in your message compares old EclipseLink with a
> draft of Mongo persistence. It would be nice to get similar numbers for
> JDBC vs. Mongo [1] using the latest code.
>
> [1] https://github.com/apache/polaris/pull/1189
>
> Thanks,
> Dmitri.
>
> On Tue, May 13, 2025 at 12:49 PM Prashant Singh
> <prashant.si...@snowflake.com.invalid> wrote:
>
> > Dear Polaris Community,
> >
> > Over the few weeks, Polaris Persistence under-went major refactor to get
> > rid of the transactional dependencies and completely adopt CAS semantics
> > (Compare and Swap), simplifying the schema.
> >
> > As a result I am pleased to announce the new JDBC implementation which is
> > considerably faster, more scalable, simpler in schema and highly
> > configurable, is now ready and a new default for the persistence layer.
> >
> > We ran the concurrency benchmarks
> > <
> >
> https://docs.google.com/document/d/1RLYaAtNUkgNW3Ef7-BWfF_8RkSK7B7oR/edit#bookmark=id.von5ayuoga6
> > >
> > conducted / contributed in April this year, with JDBC persistence and we
> > see a 100% success rate compared to roughly ~0% success rate with the
> > previous default eclipse link, with consistent p90's (< 50 ms) on mixed
> > workloads.
> >
> > You can learn more about configuration and tuning in the: Relation JDBC
> > Section <https://polaris.apache.org/in-dev/unreleased/metastores/>.
> >
> > Cheers,
> > Prashant Singh
> >
>
