Nik Samokhvalov
@samokhvalov.bsky.social
Talk to me if you have issues with your Postgres databases: [email protected] // I stand with Ukraine // Postgres.AI & DBLab & Postgres.FM
ultra-small engineering teams – 1-3 humans – might become a big trend this year
January 3, 2026 at 6:22 PM
ultra-small engineering teams – 1-3 humans – might become a big trend this year
if everyone is cooking, it's time to create awesome cookbooks
December 31, 2025 at 7:22 AM
if everyone is cooking, it's time to create awesome cookbooks
zero data loss
highest performance
always available
If you'd need to choose only 2 of these 3 for your production PostgreSQL..
What's your preference ?
highest performance
always available
If you'd need to choose only 2 of these 3 for your production PostgreSQL..
What's your preference ?
December 31, 2025 at 3:57 AM
zero data loss
highest performance
always available
If you'd need to choose only 2 of these 3 for your production PostgreSQL..
What's your preference ?
highest performance
always available
If you'd need to choose only 2 of these 3 for your production PostgreSQL..
What's your preference ?
Is the DBA role cooked?
What about DBRE -- does it suit AI era?
Who owns Postgres expertise now, in teams without DBA and without DBRE?
How do we call non-DBA engineers that grow Postgres expertise?
What tools do they need?
What about DBRE -- does it suit AI era?
Who owns Postgres expertise now, in teams without DBA and without DBRE?
How do we call non-DBA engineers that grow Postgres expertise?
What tools do they need?
December 27, 2025 at 1:04 AM
Is the DBA role cooked?
What about DBRE -- does it suit AI era?
Who owns Postgres expertise now, in teams without DBA and without DBRE?
How do we call non-DBA engineers that grow Postgres expertise?
What tools do they need?
What about DBRE -- does it suit AI era?
Who owns Postgres expertise now, in teams without DBA and without DBRE?
How do we call non-DBA engineers that grow Postgres expertise?
What tools do they need?
expert + AI > either alone
December 24, 2025 at 4:10 AM
expert + AI > either alone
If you're still manually reviewing pg_stat_statements, you're mass-producing burnout.
The best DBAs I know automated this years ago.
They focus on architecture and principles. The machine watches for problems.
That’s the gig now.
The best DBAs I know automated this years ago.
They focus on architecture and principles. The machine watches for problems.
That’s the gig now.
December 21, 2025 at 3:20 PM
If you're still manually reviewing pg_stat_statements, you're mass-producing burnout.
The best DBAs I know automated this years ago.
They focus on architecture and principles. The machine watches for problems.
That’s the gig now.
The best DBAs I know automated this years ago.
They focus on architecture and principles. The machine watches for problems.
That’s the gig now.
Building the self-driving Postgres we always wanted
December 19, 2025 at 8:41 PM
Building the self-driving Postgres we always wanted
One of the areas we adjust with almost every singe customer is reconfiguration of log_xxx parameters – this is simple and needs to be done proactively in each startup, to have data for future RCAs
defaults are not enough!
defaults are not enough!
December 15, 2025 at 7:17 PM
One of the areas we adjust with almost every singe customer is reconfiguration of log_xxx parameters – this is simple and needs to be done proactively in each startup, to have data for future RCAs
defaults are not enough!
defaults are not enough!
December 13, 2025 at 12:17 AM
BYOx is the new term (invented by claude in cursor during postgres_ai copilot development) -- it means "bring your own anything"
can be cloud (BYOC)
can be your own machine (BYOM)
can be nothing -- we'll do it for you (our cloud)
this is the way
can be cloud (BYOC)
can be your own machine (BYOM)
can be nothing -- we'll do it for you (our cloud)
this is the way
December 13, 2025 at 12:14 AM
BYOx is the new term (invented by claude in cursor during postgres_ai copilot development) -- it means "bring your own anything"
can be cloud (BYOC)
can be your own machine (BYOM)
can be nothing -- we'll do it for you (our cloud)
this is the way
can be cloud (BYOC)
can be your own machine (BYOM)
can be nothing -- we'll do it for you (our cloud)
this is the way
We went full in with monospace web
What started as local redesign of PostgresAI Console's left menu in line with copilot workflow (Monitoring → Checkup/Issues → PR/MR → Monitoring), turned out to be full redesign of the whole Console. Rolled out to all 11k+ users
And I like at a lot
dark mode WIP
What started as local redesign of PostgresAI Console's left menu in line with copilot workflow (Monitoring → Checkup/Issues → PR/MR → Monitoring), turned out to be full redesign of the whole Console. Rolled out to all 11k+ users
And I like at a lot
dark mode WIP
December 12, 2025 at 5:54 PM
We went full in with monospace web
What started as local redesign of PostgresAI Console's left menu in line with copilot workflow (Monitoring → Checkup/Issues → PR/MR → Monitoring), turned out to be full redesign of the whole Console. Rolled out to all 11k+ users
And I like at a lot
dark mode WIP
What started as local redesign of PostgresAI Console's left menu in line with copilot workflow (Monitoring → Checkup/Issues → PR/MR → Monitoring), turned out to be full redesign of the whole Console. Rolled out to all 11k+ users
And I like at a lot
dark mode WIP
among all similar screenshots you can see today, this one is out of questions
December 10, 2025 at 3:05 AM
among all similar screenshots you can see today, this one is out of questions
quick check of pg18 pgbench on my new m5 macbook pro with 10 vcpus – not bad, 0.5M select-only TPS
50k in-mem SELECTs per core is actually good -- for example 4th-gen EPYCs showed only ~10k per core postgres.fm/episodes/fou...
50k in-mem SELECTs per core is actually good -- for example 4th-gen EPYCs showed only ~10k per core postgres.fm/episodes/fou...
December 6, 2025 at 3:26 PM
quick check of pg18 pgbench on my new m5 macbook pro with 10 vcpus – not bad, 0.5M select-only TPS
50k in-mem SELECTs per core is actually good -- for example 4th-gen EPYCs showed only ~10k per core postgres.fm/episodes/fou...
50k in-mem SELECTs per core is actually good -- for example 4th-gen EPYCs showed only ~10k per core postgres.fm/episodes/fou...
switch to main and pull
December 2, 2025 at 11:05 PM
switch to main and pull
Just sent first 100 invites to PostgresAI copilot
Senior DBA on autopilot
Cyber Monday deal still open
Want in? DM me
Senior DBA on autopilot
Cyber Monday deal still open
Want in? DM me
December 1, 2025 at 9:14 PM
Just sent first 100 invites to PostgresAI copilot
Senior DBA on autopilot
Cyber Monday deal still open
Want in? DM me
Senior DBA on autopilot
Cyber Monday deal still open
Want in? DM me
"hit/read ratio" should be called "hit/(hit + read) ratio" so the math matches
or hit/nullif(hit + read, 0) to be safe
or hit/nullif(hit + read, 0) to be safe
December 1, 2025 at 6:14 AM
"hit/read ratio" should be called "hit/(hit + read) ratio" so the math matches
or hit/nullif(hit + read, 0) to be safe
or hit/nullif(hit + read, 0) to be safe
Added a new diagram to the postgres.ai front page showing how PostgresAI copilot works:
Your Postgres (located anywhere)
→ postgres_ai monitoring
→ Health check & Issues
→ GitHub PRs / AI tools
→ deployed back to PG
(Inspired by turbopuffer's clean design; it rotates 90° on mobile.)
Your Postgres (located anywhere)
→ postgres_ai monitoring
→ Health check & Issues
→ GitHub PRs / AI tools
→ deployed back to PG
(Inspired by turbopuffer's clean design; it rotates 90° on mobile.)
November 29, 2025 at 2:04 AM
Added a new diagram to the postgres.ai front page showing how PostgresAI copilot works:
Your Postgres (located anywhere)
→ postgres_ai monitoring
→ Health check & Issues
→ GitHub PRs / AI tools
→ deployed back to PG
(Inspired by turbopuffer's clean design; it rotates 90° on mobile.)
Your Postgres (located anywhere)
→ postgres_ai monitoring
→ Health check & Issues
→ GitHub PRs / AI tools
→ deployed back to PG
(Inspired by turbopuffer's clean design; it rotates 90° on mobile.)
you don't need AI to tune checkpoint_completion_target
just set it to 0.9
just set it to 0.9
November 27, 2025 at 11:10 PM
you don't need AI to tune checkpoint_completion_target
just set it to 0.9
just set it to 0.9
No PostgresFM podcast this week – was busy with trips and work, see you next week, and THANK YOU for following and listening!
November 27, 2025 at 8:40 PM
No PostgresFM podcast this week – was busy with trips and work, see you next week, and THANK YOU for following and listening!
Tools show green "CPU" when it's actually I/O. And COPY is what pg_dump and pg_restore use by default.
We went in to add wait events for COPY TO/FROM PROGRAM, then realized the gap is much bigger than expected. Verified with pg_wait_sampling.
1/
We went in to add wait events for COPY TO/FROM PROGRAM, then realized the gap is much bigger than expected. Verified with pg_wait_sampling.
1/
PostgresTV
Welcome to Postgres TV, a place where we discuss all things PostgreSQL.
For inquiries: [email protected]
postgres.tv
November 26, 2025 at 11:00 PM
Tools show green "CPU" when it's actually I/O. And COPY is what pg_dump and pg_restore use by default.
We went in to add wait events for COPY TO/FROM PROGRAM, then realized the gap is much bigger than expected. Verified with pg_wait_sampling.
1/
We went in to add wait events for COPY TO/FROM PROGRAM, then realized the gap is much bigger than expected. Verified with pg_wait_sampling.
1/
Hacking Postgres online – join! www.youtube.com/watch?v=wLPD...
Hacking Postgres with Andrey, Kirk, and Nik
YouTube video by PostgresTV
www.youtube.com
November 26, 2025 at 5:42 PM
Hacking Postgres online – join! www.youtube.com/watch?v=wLPD...
Asked AI to analyze what wait events are missing in current PostgreSQL code – after a few iterations, accumulated > 70 gaps, here is the report: github.com/NikolayS/pos...
Checking with -hackers if it makes sense to dig further: www.postgresql.org/message-id/f...
Checking with -hackers if it makes sense to dig further: www.postgresql.org/message-id/f...
github.com
November 22, 2025 at 12:47 AM
Asked AI to analyze what wait events are missing in current PostgreSQL code – after a few iterations, accumulated > 70 gaps, here is the report: github.com/NikolayS/pos...
Checking with -hackers if it makes sense to dig further: www.postgresql.org/message-id/f...
Checking with -hackers if it makes sense to dig further: www.postgresql.org/message-id/f...
I like how easy it is now to get the new postgres.ai copilot for any PostgreSQL:
- One-click deployment in three modes – PaaS, BYOC, BYOM
- Easy access to dashboards at xxxxxx.pgai.watch by default – to dive into all the details
- "Resolve in Cursor" (or you can use Claude Code, Codex, etc.)
- One-click deployment in three modes – PaaS, BYOC, BYOM
- Easy access to dashboards at xxxxxx.pgai.watch by default – to dive into all the details
- "Resolve in Cursor" (or you can use Claude Code, Codex, etc.)
November 14, 2025 at 4:34 AM
I like how easy it is now to get the new postgres.ai copilot for any PostgreSQL:
- One-click deployment in three modes – PaaS, BYOC, BYOM
- Easy access to dashboards at xxxxxx.pgai.watch by default – to dive into all the details
- "Resolve in Cursor" (or you can use Claude Code, Codex, etc.)
- One-click deployment in three modes – PaaS, BYOC, BYOM
- Easy access to dashboards at xxxxxx.pgai.watch by default – to dive into all the details
- "Resolve in Cursor" (or you can use Claude Code, Codex, etc.)
November 14, 2025 at 1:49 AM