How to calculate PostgreSQL memory usage on Linux?

Question

I searched a lot about how to calculate memory usage of PostgreSQL process on Linux. I read article about how to calculate Memory usage for a generic process but I think that PostgreSQL has some peculiarity. For example, it has some basic processes:logger, checkpointer, bg writer, etc. But Linux creates also a process for each client connection on the master node.

The easy way to calculate the memory usage is with ps command listing the RSS of each process:

ps -aux | grep -v grep | grep postgres | awk '{ print $6 }'

and then sum it. But this doesn't work since the result is larger than the total memory. Some articles suggest the use of:

/proc/PID/smaps

but as said above I have more PID and I am unable to find a website that shows a script that let me easily calculate this information.

I found this interesting article, but it's not clear to me how to convert it into a working script. https://www.depesz.com/2012/06/09/how-much-ram-is-postgresql-using/

Does anyone know which is the best approach to solve this issue?

Stack Overflow is for programming questions. You might try asking on Unix & Linux, Server Fault or Super User instead. — John Kugelman
– John Kugelman, Commented Dec 15, 2020 at 18:07
Very strange. Search "linux memory usage" you'll find thousands of questions like mine, unfortunately, I haven't found an answer. However, I will follow your suggestion. However, my final request is about a working script that can be considered "programming" in some way. — Salvatore D'angelo
– Salvatore D'angelo, Commented Dec 15, 2020 at 18:12

Laurenz Albe · Accepted Answer · 2020-12-15 18:49:34Z

3

To apply the information from the blog you quote:

ps -u postgres o pid= | \
sed 's# *\(.*\)#/proc/\1/smaps#' | \
xargs sudo grep ^Pss: | \
awk '{A+=$2} END{print A}'

first, get the process numbers of all processes running under user postgres
then get the name of the corresponding smaps file (/proc/PID/smaps)
then get the Pss line from that file, which contains the “proportional stack size” which divides shared memory by the number of processes attached to it
finally, add up the numbers

answered Dec 15, 2020 at 18:49

Laurenz Albe

257k22 gold badges312 silver badges388 bronze badges

Sign up to request clarification or add additional context in comments.

7 Comments

Salvatore D'angelo Over a year ago

Only one doubt. The original article says: "Basically, Pss is at most Rss, but gets decreased if the same pages of memory are used by more than one process." so PSS is (RSS-common memory), but my point is that common memory should be count at least once and I am not sure this occurs. Am I correct?

Salvatore D'angelo Over a year ago

I mean, shared memory should be count at least once.

Laurenz Albe Over a year ago

Correct. If there is only one process mapping the shared memory segment, PSS = RSS for that segment. If there are more, the value gets divided by the number of processes. So if you calculate the sum, you will end up with the correct number.

Salvatore D'angelo Over a year ago

I calculated RSS and PSS for each PostgreSQL process on a slave node and I don't see any process where PSS=RSS. So I assume I have to add it in some way. Any suggestion? docs.google.com/document/d/…

Laurenz Albe Over a year ago

Even on a standby server there are always some background processes, like the checkpointer and the background writer (and of course the WAL receiver). They all attach to and use shared memory.

|

Collectives™ on Stack Overflow

How to calculate PostgreSQL memory usage on Linux?

1 Answer 1

7 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

7 Comments

Your Answer

Sign up or log in

Post as a guest

Related