How do I update 1.3 billion rows in this table more efficiently? - postgresql

Question

I have 1.3 billion rows in a PostgreSQL table sku_comparison that looks like this:
id1 (INTEGER) | id2 (INTEGER) | (10 SMALLINT columns) | length1 (SMALLINT)... |
... length2 (SMALLINT) | length_difference (SMALLINT)
The id1 and id2 columns are referenced in a table called sku, which contains about 300,000 rows, and have an associated varchar(25) value in each row from a column, code.
There is a btree index built on id1 and id2, and a compound index of id1 and id2 in sku_comparison. There is a btree index on the id column of sku, as well.
My goal is to update the length1 and length2 columns with the lengths of the corresponding code column from the sku table. However, I ran the following code for over 20 hours, and it did not complete the update:
UPDATE sku_comparison SET length1=length(sku.code) FROM sku
WHERE sku_comparison.id1=sku.id;
All of the data is stored on a single hard disk on a local computer, and the processor is fairly modern. Constructing this table, which required much more complicated string comparisons in Python, only took about 30 hours or so, so I am not sure why something like this would take as long.
edit: here are formatted table definitions:
Table "public.sku"
Column | Type | Modifiers
------------+-----------------------+--------------------------------------------------
id | integer | not null default nextval('sku_id_seq'::regclass)
sku | character varying(25) |
pattern | character varying(25) |
pattern_an | character varying(25) |
firsttwo | character(2) | default ' '::bpchar
reference | character varying(25) |
Indexes:
"sku_pkey" PRIMARY KEY, btree (id)
"sku_sku_idx" UNIQUE, btree (sku)
"sku_firstwo_idx" btree (firsttwo)
Referenced by:
TABLE "sku_comparison" CONSTRAINT "sku_comparison_id1_fkey" FOREIGN KEY (id1) REFERENCES sku(id)
TABLE "sku_comparison" CONSTRAINT "sku_comparison_id2_fkey" FOREIGN KEY (id2) REFERENCES sku(id)
Table "public.sku_comparison"
Column | Type | Modifiers
---------------------------+----------+-------------------------
id1 | integer | not null
id2 | integer | not null
consec_charmatch | smallint |
consec_groupmatch | smallint |
consec_fieldtypematch | smallint |
consec_groupmatch_an | smallint |
consec_fieldtypematch_an | smallint |
general_charmatch | smallint |
general_groupmatch | smallint |
general_fieldtypematch | smallint |
general_groupmatch_an | smallint |
general_fieldtypematch_an | smallint |
length1 | smallint | default 0
length2 | smallint | default 0
length_difference | smallint | default '-999'::integer
Indexes:
"sku_comparison_pkey" PRIMARY KEY, btree (id1, id2)
"ssd_id1_idx" btree (id1)
"ssd_id2_idx" btree (id2)
Foreign-key constraints:
"sku_comparison_id1_fkey" FOREIGN KEY (id1) REFERENCES sku(id)
"sku_comparison_id2_fkey" FOREIGN KEY (id2) REFERENCES sku(id)

Answer 1

Would you consider using an anonymous code block?
using pseudo code...
FOREACH 'SELECT ski.id,
sku.code,
length(sku.code)
FROM sku
INTO v_skuid, v_skucode, v_skulength'
DO
UPDATE sku_comparison
SET sku_comparison.length1 = v_skulength
WHERE sku_comparison.id1=v_skuid;
END DO
END FOREACH
This would break the whole thing into smaller transactions and you will not be evaluating the length of sku.code every time.