07-27-2023, 11:18 AM
One very late contribution... in case it helps anyone waaaaaay down the line... I had a task to find matching pairs of transactions (actually both sides of account-to-account transfers) in a banking app, to identify which ones were the 'from' and 'to' for each inter-account-transfer transaction, so we ended up with this:
SELECT
LEAST(primaryid, secondaryid) AS transactionid1,
GREATEST(primaryid, secondaryid) AS transactionid2
FROM (
SELECT table1.transactionid AS primaryid,
table2.transactionid AS secondaryid
FROM financial_transactions table1
INNER JOIN financial_transactions table2
ON table1.accountid = table2.accountid
AND table1.transactionid <> table2.transactionid
AND table1.transactiondate = table2.transactiondate
AND table1.sourceref = table2.destinationref
AND table1.amount = (0 - table2.amount)
) AS DuplicateResultsTable
GROUP BY transactionid1
ORDER BY transactionid1;
The result is that the `DuplicateResultsTable` provides rows containing matching (i.e. duplicate) transactions, but it also provides the same transaction id's in reverse the second time it matches the same pair, so the outer `SELECT` is there to group by the first transaction ID, which is done by using `LEAST` and `GREATEST` to make sure the two transactionid's are always in the same order in the results, which makes it safe to `GROUP` by the first one, thus eliminating all the duplicate matches. Ran through nearly a million records and identified 12,000+ matches in just under 2 seconds. Of course the transactionid is the primary index, which really helped.
SELECT
LEAST(primaryid, secondaryid) AS transactionid1,
GREATEST(primaryid, secondaryid) AS transactionid2
FROM (
SELECT table1.transactionid AS primaryid,
table2.transactionid AS secondaryid
FROM financial_transactions table1
INNER JOIN financial_transactions table2
ON table1.accountid = table2.accountid
AND table1.transactionid <> table2.transactionid
AND table1.transactiondate = table2.transactiondate
AND table1.sourceref = table2.destinationref
AND table1.amount = (0 - table2.amount)
) AS DuplicateResultsTable
GROUP BY transactionid1
ORDER BY transactionid1;
The result is that the `DuplicateResultsTable` provides rows containing matching (i.e. duplicate) transactions, but it also provides the same transaction id's in reverse the second time it matches the same pair, so the outer `SELECT` is there to group by the first transaction ID, which is done by using `LEAST` and `GREATEST` to make sure the two transactionid's are always in the same order in the results, which makes it safe to `GROUP` by the first one, thus eliminating all the duplicate matches. Ran through nearly a million records and identified 12,000+ matches in just under 2 seconds. Of course the transactionid is the primary index, which really helped.