Обсуждение: BUG #16443: Too much memory usage on insert query

The following bug has been logged on the website:

Bug reference:      16443
Logged by:          Kurt Roeckx
Email address:      kurt@roeckx.be
PostgreSQL version: 12.3
Operating system:   Debian
Description:

Hi,

I have this in my config file:
shared_buffers = 2048MB                 # min 128kB
work_mem = 1024MB                       # min 64kB

I was executing this query:
insert into ct_entry_chain (ct_entry_id, certificate_id) select id,
unnest(certificate_chain_id) from ct_entry;

The process was using at least 14 GB, of the 8 GB of RAM that's available.
Swap was also using around the 8GB maximum, while normally around 200 MB is
swapped out. I didn't expect this process to use more than around 3 GB.

This has resulted in an out of memory condition.

I was converting my schema, ct_entry_chain is a new table that looks like:
                                Table "public.ct_entry_chain"
     Column     |  Type  | Collation | Nullable |                  Default
----------------+--------+-----------+----------+--------------------------------------------
 id             | bigint |           | not null |
nextval('ct_entry_chain_id_seq'::regclass)
 ct_entry_id    | bigint |           | not null |
 certificate_id | bigint |           | not null |
Indexes:
    "ct_entry_chain_pkey" PRIMARY KEY, btree (id)
    "ct_entry_chain_ct_entry_id_certificate_id_key" UNIQUE CONSTRAINT, btree
(ct_entry_id, certificate_id)
Foreign-key constraints:
    "ct_entry_chain_certificate_id_fkey" FOREIGN KEY (certificate_id)
REFERENCES raw_certificates(id)
    "ct_entry_chain_ct_entry_id_fkey" FOREIGN KEY (ct_entry_id) REFERENCES
ct_entry(id)

ct_entry itself looks like:
                                            Table "public.ct_entry"
        Column        |           Type           | Collation | Nullable |
           Default
----------------------+--------------------------+-----------+----------+--------------------------------------
 id                   | bigint                   |           | not null |
nextval('ct_entry_id_seq'::regclass)
 ct_log_id            | integer                  |           | not null |
 raw_certificate_id   | bigint                   |           | not null |
 log_entry            | bigint                   |           | not null |
 timestamp            | timestamp with time zone |           | not null |
 version              | integer                  |           | not null |
 entry_type           | integer                  |           | not null |
 leaf_type            | integer                  |           | not null |
 extensions           | bytea                    |           |          |
 issuer_key_hash      | bytea                    |           |          |
 pre_certificate_id   | bigint                   |           |          |
 certificate_chain_id | bigint[]                 |           |          |
Indexes:
    "ct_entry_pkey" PRIMARY KEY, btree (id)
    "ct_entry_raw_certificate_id_idx" btree (raw_certificate_id)
Foreign-key constraints:
    "ct_entry_ct_log_id_fkey" FOREIGN KEY (ct_log_id) REFERENCES
ct_logs(id)
    "ct_entry_pre_certificate_id_fkey" FOREIGN KEY (pre_certificate_id)
REFERENCES raw_certificates(id)
    "ct_entry_raw_certificate_id_fkey" FOREIGN KEY (raw_certificate_id)
REFERENCES raw_certificates(id)
Referenced by:
    TABLE "ct_entry_chain" CONSTRAINT "ct_entry_chain_ct_entry_id_fkey"
FOREIGN KEY (ct_entry_id) REFERENCES ct_entry(id)

If I retry the query, I see the memory grow slowly. If I
cancel the query, memory returns to normal.

PG Bug reporting form <noreply@postgresql.org> writes:
> I was executing this query:
> insert into ct_entry_chain (ct_entry_id, certificate_id) select id,
> unnest(certificate_chain_id) from ct_entry;

How much data is that?

> The process was using at least 14 GB, of the 8 GB of RAM that's available.

My first guess is that the space was being eaten by trigger list entries
to verify the foreign-key constraints on the target table.  You might be
better advised to fill the new table first and then create its FK
constraints.  (Building the indexes afterwards wouldn't be a bad idea,
either.)

We are looking at better mechanisms for handling FK verification, but
that won't see the light of day before v14 at the earliest.

            regards, tom lane

On Sun, May 17, 2020 at 11:40:53AM -0400, Tom Lane wrote:
> PG Bug reporting form <noreply@postgresql.org> writes:
> > I was executing this query:
> > insert into ct_entry_chain (ct_entry_id, certificate_id) select id,
> > unnest(certificate_chain_id) from ct_entry;
> 
> How much data is that?

The ct_entry table contains 305 GB data over about 2.1e9 rows.
certificate_chain_id contains about 2.2 entries per row.

> > The process was using at least 14 GB, of the 8 GB of RAM that's available.
> 
> My first guess is that the space was being eaten by trigger list entries
> to verify the foreign-key constraints on the target table.  You might be
> better advised to fill the new table first and then create its FK
> constraints.  (Building the indexes afterwards wouldn't be a bad idea,
> either.)
> 
> We are looking at better mechanisms for handling FK verification, but
> that won't see the light of day before v14 at the earliest.

I will try that, thanks.


Kurt