Get size of large object in PostgreSQL query?

Question

I would like to obtain the byte size of a blob.

I am using Postgresql and would like to obtain the size using an SQL query. Something like this:

SELECT sizeof(field) FROM table;

Is this possible in Postgresql?

Update: I have read the postgresql manual and could not find an appropriate function to calculate the file size. Also, the blob is stored as a large object.

Solution 1

Not that I've used large objects, but looking at the docs: http://www.postgresql.org/docs/current/interactive/lo-interfaces.html#LO-TELL

I think you have to use the same technique as some file system APIs require: seek to the end, then tell the position. PostgreSQL has SQL functions that appear to wrap the internal C functions. I couldn't find much documentation, but this worked:

CREATE OR REPLACE FUNCTION get_lo_size(oid) RETURNS bigint
VOLATILE STRICT
LANGUAGE 'plpgsql'
AS $$
DECLARE
    fd integer;
    sz bigint;
BEGIN
    -- Open the LO; N.B. it needs to be in a transaction otherwise it will close immediately.
    -- Luckily a function invocation makes its own transaction if necessary.
    -- The mode x'40000'::int corresponds to the PostgreSQL LO mode INV_READ = 0x40000.
    fd := lo_open($1, x'40000'::int);
    -- Seek to the end.  2 = SEEK_END.
    PERFORM lo_lseek(fd, 0, 2);
    -- Fetch the current file position; since we're at the end, this is the size.
    sz := lo_tell(fd);
    -- Remember to close it, since the function may be called as part of a larger transaction.
    PERFORM lo_close(fd);
    -- Return the size.
    RETURN sz;
END;
$$; 

Testing it:

-- Make a new LO, returns an OID e.g. 1234567
SELECT lo_create(0);

-- Populate it with data somehow
...

-- Get the length.
SELECT get_lo_size(1234567);

It seems the LO functionality is designed to be used mostly through the client or through low-level server programming, but at least they've provided some SQL visible functions for it, which makes the above possible. I did a query for SELECT relname FROM pg_proc where relname LIKE 'lo%' to get myself started. Vague memories of C programming and a bit of research for the mode x'40000'::int and SEEK_END = 2 value were needed for the rest!

Solution 2

You could change your application to store the size when you create the large object. Otherwise you can use a query such as:

select sum(length(lo.data)) from pg_largeobject lo
where lo.loid=XXXXXX

You can use also the large object API functions, as suggested in a previous post, they work ok, but are an order of magnitude slower than the select method suggested above.

Solution 3

select pg_column_size(lo_get(lo_oid)) from table;

Gives you the size in bytes.

If you want pretty printing:

select pg_size_pretty(pg_column_size(lo_get(lo_oid))::numeric) from table;

Solution 4

Try length() or octet_length()

Solution 5

This is my solution:

select
lo.loid,
pg_size_pretty(sum(octet_length(lo.data)))
from pg_largeobject lo
where lo.loid in (select pg_largeobject.loid from pg_largeobject)
group by lo.loid;

View more solutions

Comments

[MarekM]

This is the cleanest solution. Thanks

[ochedru]

Unfortunately, the pg_largeobject catalog is no longer publicly accessible since PostgreSQL 9.0: postgresql.org/docs/current/static/catalog-pg-largeobject.ht‌​ml.

[wutzebaer]

somehow it is always 4 bytes more than it actually is, why?

[ochedru]

To avoid "result out of range" errors and make it work with large objects larger than 2GB, use lo_seek64 and lo_tell64.

[OrangeDog]

@wutzebaer the first four bytes of a bytea column are the size of the rest of it

[OrangeDog]

How does the performance compare to the other answers?

[OrangeDog]

i.e. you should use octet_length (as per other answers) not pg_column_size