Tuning PostgreSQL for High Write Workloads

Tuning PostgreSQL for High Write Workloads - Grant McAlister

[@AlexeyMatushevsky]

Great talk! thank you.

[@dengan699]

Very informative!

[@2008topshelf]

Good job Grant!!

[@nareshgb1]

really gets into the details.

[@BestTechLearn]

Informative..

[@piyushkatariya1040]

wow. I hope some of these features get implemented in Postgres core

[@jayextarys8616]

Postgres is getting better and better but adding huge neat features takes so much time :(

[@renatocron]

Well, you see, the main problem here was the FPW, and it only there to be safe to use it at hardware configurations. Changing update to other methods where FPW is not necessary kind hard, but maybe possible, but that's not solved right now (even with PG12, now we have some work done on the internal access API, but that's only the beginning to allow different storage engines). Some indexes already try to avoid constant re-balancing (GIN with fastupdates) and the same technique would need to be applied to get rid of FPW on any hardware.

[@0911jedi]

Thanks for the very informative video. We are victim of GUID based B-Tree indexes. I have following queries 1. Can Hash Index help for GUID (random numbers)? 2. Can Arora help in solving B-Tree performance issues or we need to rely on right side generation of keys for B-Tree?

[@emiliocolon3245]

That is but when one does not optimize the postgresql 25,000 TPS, but optimized for 4MB to 4GB it gives 443000 TPS

[@IvanMushketik]

21:08 Would using UUID Version 1 instead of random UUID help to constrain randomness?

[@dharmaturtle]

No. v1 UUIDs are not lexicographically sortable. You can see this if you go to any online generator and click generate a few times - you'll notice that the first part of the hex string changes far more than the next two sections. This is why Sql Server has NEWSEQUENTIALID, which doesn't follow RFC 4122 (the UUID spec). For PostGres, I've been recommending that people use ULIDs (Universally Unique Lexicographically Sortable Identifier), which you should google (not linking for fear of antispam).

[@user-gi4qu9do2v]

there is guidv7. first 32 bits is timestamp

[@marcosoliveira8731]

Really good talk. Is aurora downloadble ?

[@MrBenoitL]

no sir it relies on the AWS Infrastructure to do all these neat things.

[@jayextarys8616]

Proprietary stuff... GRR

[@krishanjangid212]

Hi, How can I generate these statistics reports while running the benchmarks?

[@VivekSingh06]

pg_bench

[@krishanjangid212]

Vivek Singh what I'm looking for, is an automated way that generates statistics reports after each run of pgbench. So far, I've not been able to find such method. Can you please let me know of there is one in detail? Thanks

[@123wlpha2]

[/dog][/dog][/dog] Optimization tips using cloud databases Cloud Database > Reduce Indexes > Non-random Data > WAL Buffer size > WAL Compression > No Optimization

[@luigic6544]

someone who can help me to make an inheritance in my shell postgesql please!!!!