Thread: Memory footprint diff between 9.5 and 12

On Thu, 2020-05-07 at 13:33 -0700, Tory M Blue wrote:
> I hadn't noticed this until today, but a running 9.5 system with buffers at 10GB starts
> and has been running years without issues. (15GB available)
>
> Postgres 12 will not start with that configuration, complaining about memory availability.
> So Postgres12 won't start until shared buffers is 6GB, but even with that, my DB servers ,
> postgres queries started complaining about being unable to allocate memory "unable to allocate".
>
> So dropping them to 4GB (on a 15GB system), may help, but did I miss a huge note about
> significant memory changes between 9.5 to 12?
>
> Is there something else I'm missing that on busy systems is important, something introduced
> in 10 or 11 as again I'm not seeing anything noted in 12.

There must be something else running on the machine that allocates memory.

Did you perchance run the 9.5 and the v12 server on the same machine?

Yours,
Laurenz Albe
--

Tory M Blue <tmblue@gmail.com> writes:
> 12 will not start at 10GB, even though it's the same hardware, same config
> file, same physical box, same everything, just version 12 vs 9.5

For me, using all-default settings (in particular, shared_buffers =
128MB), the shared memory block is about 141.6MB using 9.5 and 142.1MB
using 12.  So there's half a meg or so of additional data in v12, but
certainly not gigabytes worth.

Are you trying to start both postmasters concurrently?  Maybe you're
hitting some kernel limit on the total amount of shared memory in the
system.

                        regards, tom lane

Tory M Blue <tmblue@gmail.com> writes:
> 12 will not start at 10GB, even though it's the same hardware, same config
> file, same physical box, same everything, just version 12 vs 9.5

For me, using all-default settings (in particular, shared_buffers =
128MB), the shared memory block is about 141.6MB using 9.5 and 142.1MB
using 12.  So there's half a meg or so of additional data in v12, but
certainly not gigabytes worth.

Are you trying to start both postmasters concurrently?  Maybe you're
hitting some kernel limit on the total amount of shared memory in the
system.

                        regards, tom lane

shared_memory_type (enum)

Specifies the shared memory implementation that the server should use for the main shared memory region that holds PostgreSQL's shared buffers and other shared data. Possible values are mmap (for anonymous shared memory allocated using mmap), sysv (for System V shared memory allocated via shmget) and windows (for Windows shared memory). Not all values are supported on all platforms; the first supported option is the default for that platform. The use of the sysv option, which is not the default on any platform, is generally discouraged because it typically requires non-default kernel settings to allow for large allocations (see Section 18.4.1).

 

On Mon, May 11, 2020 at 1:36 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:

Tory M Blue <tmblue@gmail.com> writes:
> 12 will not start at 10GB, even though it's the same hardware, same config
> file, same physical box, same everything, just version 12 vs 9.5

For me, using all-default settings (in particular, shared_buffers =
128MB), the shared memory block is about 141.6MB using 9.5 and 142.1MB
using 12.  So there's half a meg or so of additional data in v12, but
certainly not gigabytes worth.

Are you trying to start both postmasters concurrently?  Maybe you're
hitting some kernel limit on the total amount of shared memory in the
system.

                        regards, tom lane

Okay the one difference I see in settings is this little gem in 12.. 

shared_memory_type

mmap

Otherwise i'm not seeing a ton of other settings not common between them,.

This is the only major difference I'm seeing, as it's really not an option in 9.5..... Appears 9.5 was using

shared_memory_type (enum)

Specifies the shared memory implementation that the server should use for the main shared memory region that holds PostgreSQL's shared buffers and other shared data. Possible values are mmap (for anonymous shared memory allocated using mmap), sysv (for System V shared memory allocated via shmget) and windows (for Windows shared memory). Not all values are supported on all platforms; the first supported option is the default for that platform. The use of the sysv option, which is not the default on any platform, is generally discouraged because it typically requires non-default kernel settings to allow for large allocations (see Section 18.4.1).

 

On Monday, May 11, 2020, Tory M Blue <tmblue@gmail.com> wrote:

I'll keep digging, but i'm hesitant to do my multiple TB db's with half of their shared buffer configs, until I understand what 12 is doing differently than 9.5

Maybe run your test suite on 9.6, 10, and 11 to see if it is indeed new to 12 or at least appears on other versions?

David J. 

On Tue, May 12, 2020 at 2:52 PM Tory M Blue <tmblue@gmail.com> wrote:
> It took the change but didn't help. So 10GB of shared_buffers in 12 is still a no go. I'm down to 5GB and it works, but this is the same hardware, the same exact 9.5 configuration. So I'm missing something. WE have not had to mess with kernel memory settings since 9.4, so this is an odd one.
>
> I'll keep digging, but i'm hesitant to do my multiple TB db's with half of their shared buffer configs, until I understand what 12 is doing differently than 9.5

Which exact version of 9.5.x are you coming from?  What's the exact
error message on 12 (you showed the shared_memory_type=sysv error, but
with the default  value (mmap) how does it look)?  What's your
huge_pages setting?

Can you reproduce the problem with a freshly created test cluster?  As
a regular user, assuming regular RHEL packaging, something like
/usr/pgsql-12/bin/initdb -D test_pgdata, and then
/usr/pgsql-12/bin/postgres -D test_pgdata -c shared_buffers=10GB (then
^C to stop it).  If that fails to start in the same way, it'd be
interesting to see the output of the second command with strace in
front of it, in the part where it allocates shared memory.  And
perhaps it'd be interesting to see the same output with
/usr/pgsql-9.5/bin/XXX (if you still have the packages).  For example,
on my random dev laptop that looks like:

openat(AT_FDCWD, "/proc/meminfo", O_RDONLY) = 6
fstat(6, {st_mode=S_IFREG|0444, st_size=0, ...}) = 0
read(6, "MemTotal:       16178852 kB\nMemF"..., 1024) = 1024
read(6, ":    903168 kB\nShmemHugePages:  "..., 1024) = 311
close(6)                                = 0
mmap(NULL, 11016339456, PROT_READ|PROT_WRITE,
MAP_SHARED|MAP_ANONYMOUS|MAP_HUGETLB, -1, 0) = -1 ENOMEM (Cannot
allocate memory)
mmap(NULL, 11016003584, PROT_READ|PROT_WRITE,
MAP_SHARED|MAP_ANONYMOUS, -1, 0) = 0x7ff74e579000
shmget(0x52e2c1, 56, IPC_CREAT|IPC_EXCL|0600) = 3244038
shmat(3244038, NULL, 0)                 = 0x7ff9df5ad000

The output is about the same on REL9_5_STABLE and REL_12_STABLE for
me, only slightly different sizes.  If that doesn't fail in the same
way on your system with 12, perhaps there are some more settings from
your real clusters required to make it fail.  You could add them one
by one with -c foo=bar or in the throw away
test_pgdata/postgresql.conf, and perhaps that process might shed some
light?

I was going to ask if it might be a preloaded extension that is asking
for gobs of extra memory in 12, but we can see from your "Failed
system call was shmget(key=5432001, size=11026235392, 03600)" that
it's in the same ballpark as my total above for shared_buffers=10GB.

Tory M Blue <tmblue@gmail.com> writes:
> That may be the next step in the lab, but was hoping someone knew of a
> significant difference.

I think we've made it perfectly clear that we don't.  There's something
odd about your situation.

                        regards, tom lane

On Monday, May 11, 2020, David G. Johnston <david.g.johnston@gmail.com> wrote:

Repost, edited subject by mistake...

On Monday, May 11, 2020, Tory M Blue <tmblue@gmail.com> wrote:

And just to repeat. Same exact hardware, same kernel, nothing more than installing the latest postgres12, copying my config files from 9.5 to 12 and running the pg_upgrade.

You’ll want to remove the pg_upgrade from the equation and try v12 

Sorry...if you copied the config to v12 before the upgrade and the upgrade worked that suggests that v12 booted up at some point with the configuration, no?  Does pg_upgrade do something special?

David J

On Mon, May 11, 2020 at 9:01 PM Thomas Munro <thomas.munro@gmail.com> wrote:

On Tue, May 12, 2020 at 2:52 PM Tory M Blue <tmblue@gmail.com> wrote:
> It took the change but didn't help. So 10GB of shared_buffers in 12 is still a no go. I'm down to 5GB and it works, but this is the same hardware, the same exact 9.5 configuration. So I'm missing something. WE have not had to mess with kernel memory settings since 9.4, so this is an odd one.
>
> I'll keep digging, but i'm hesitant to do my multiple TB db's with half of their shared buffer configs, until I understand what 12 is doing differently than 9.5

Which exact version of 9.5.x are you coming from?  What's the exact
error message on 12 (you showed the shared_memory_type=sysv error, but
with the default  value (mmap) how does it look)?  What's your
huge_pages setting?

9.5-20

postgresql95-9.5.20-2PGDG.rhel7.x86_64
postgresql95-contrib-9.5.20-2PGDG.rhel7.x86_64
postgresql95-libs-9.5.20-2PGDG.rhel7.x86_64

postgresql95-server-9.5.20-2PGDG.rhel7.x86_64 

I don't use huge_pages

And this error is actually from the default mmap

May 08 12:33:58 qdb01.prod.ca postmaster[8790]: < 2020-05-08 12:33:58.324 PDT  >HINT:  This error usually means that PostgreSQL's request for a shared memory segment exceeded available memory, swap space, or huge pages. To reduce the request size (currently 11026235392 bytes), reduce PostgreSQL's shared memory usage, perhaps by reducing shared_buffers or max_connections.

The above error is with 12 trying to start with shared_buffers = 10GB...

9.5 starts fine with the same configuration file.   That kind of started me down this path.

And just to repeat. Same exact hardware, same kernel, nothing more than installing the latest postgres12, copying my config files from 9.5 to 12 and running the pg_upgrade.

9.5 has been running for years with the same configuration file, so something changed somewhere along the line that is preventing 12 to start with the same config file.  And the allocation error is with either the sysv or mman on 12. (will start with 5GB allocated, but not 10GB, on a 15GB box (dedicated postgres server).

Can you reproduce the problem with a freshly created test cluster?  As
a regular user, assuming regular RHEL packaging, something like
/usr/pgsql-12/bin/initdb -D test_pgdata, and then
/usr/pgsql-12/bin/postgres -D test_pgdata -c shared_buffers=10GB (then
^C to stop it).  If that fails to start in the same way, it'd be
interesting to see the output of the second command with strace in
front of it, in the part where it allocates shared memory.  And
perhaps it'd be interesting to see the same output with
/usr/pgsql-9.5/bin/XXX (if you still have the packages).  For example,
on my random dev laptop that looks like:

openat(AT_FDCWD, "/proc/meminfo", O_RDONLY) = 6
fstat(6, {st_mode=S_IFREG|0444, st_size=0, ...}) = 0
read(6, "MemTotal:       16178852 kB\nMemF"..., 1024) = 1024
read(6, ":    903168 kB\nShmemHugePages:  "..., 1024) = 311
close(6)                                = 0
mmap(NULL, 11016339456, PROT_READ|PROT_WRITE,
MAP_SHARED|MAP_ANONYMOUS|MAP_HUGETLB, -1, 0) = -1 ENOMEM (Cannot
allocate memory)
mmap(NULL, 11016003584, PROT_READ|PROT_WRITE,
MAP_SHARED|MAP_ANONYMOUS, -1, 0) = 0x7ff74e579000
shmget(0x52e2c1, 56, IPC_CREAT|IPC_EXCL|0600) = 3244038
shmat(3244038, NULL, 0)                 = 0x7ff9df5ad000

The output is about the same on REL9_5_STABLE and REL_12_STABLE for
me, only slightly different sizes.  If that doesn't fail in the same
way on your system with 12, perhaps there are some more settings from
your real clusters required to make it fail.  You could add them one
by one with -c foo=bar or in the throw away
test_pgdata/postgresql.conf, and perhaps that process might shed some
light?

I was going to ask if it might be a preloaded extension that is asking
for gobs of extra memory in 12, but we can see from your "Failed
system call was shmget(key=5432001, size=11026235392, 03600)" that
it's in the same ballpark as my total above for shared_buffers=10GB.

Be more than happy to test this out. I'll see what I can pull tomorrow and provide some dataz :)   I know it's not ideal to use the same config file, I know that various things are added or changed (usually added) but the defaults are typically safe. But after sometime dialing in the settings for our use case, I've just kind of kept moving them forward.

But  let me do some more testing tomorrow (since I'm trying to get to the bottom of this, before I attempt my big DB upgrades).  So I'll spend some time testing and see if I can't get similar "failures/challenges"? and go from there.

Appreciate the ideas!

Tory