Quota exceeded: Your project exceeded quota for free query bytes scanned

You used the 1 TiB of free monthly query processing and the account has no paid billing attached. Nothing you change in the query fixes it. Here is what counts against the TiB and what does not.

Your project exceeded quota for free query bytes scanned

What BigQuery means

The free usage tier gives each billing account 1 TiB of query processing per month. You have used it, and the account has no paid Cloud Billing attached, so there is nothing to charge the next byte to.

This is not a rate limit. Waiting a few minutes does nothing. Rewriting the query does nothing, because even a cheap query needs some budget and there is none.

Two ways out, and they are the only two: wait for the calendar month to roll over, or upgrade the account to paid Cloud Billing. Google’s resolution for this error is the upgrade.

Who sees it

Mostly people in the BigQuery sandbox, which is BigQuery without a credit card. It is the same 1 TiB. A single SELECT * over a wide public table can be a large fraction of it.

If you are on a paid account you will not see this message. You will see a bill, or you will hit maximum_bytes_billed or a custom quota, which are different errors.

What counts against the 1 TiB

The number is bytes processed by query jobs, which is not the same as the size of your tables or the size of your results.

Charged:

• the full uncompressed size of every column a query reads, for the partitions it reads

• the same bytes again on every re-run, unless the result came from cache

• metadata. An INFORMATION_SCHEMA query is a query job. It bills a minimum of 10 MiB even when it reads less, and it is never served from cache, so every run costs again.

Not charged:

• a cache hit. The same query text against unchanged tables reads nothing. See Query cache.
• a dry run. Validating a query is free, however many times you do it.
• batch loading, copying a table within one region, exporting with an extract job, and deleting datasets, tables, partitions, views or functions. Those are free operations. The EXPORT DATA SQL statement is a query, so that one is billed.

Storage is a separate free allowance, 10 GiB per month, with its own meter.

Making the TiB go further

• Name your columns. BigQuery is columnar, so SELECT * on a 200-column table reads 200 columns. SELECT * EXCEPT (big_json_blob) is already a large saving. This is the single biggest lever.
• Filter on the partitioning column. A WHERE on a partitioned table’s partition column skips whole partitions. A WHERE on any other column reads everything and then discards. See Partitioning.
• LIMIT does not reduce what is read. SELECT * FROM t LIMIT 10 on an unclustered table scans the whole table. It is the most common misunderstanding about BigQuery billing.
• Preview instead of querying. The table preview in the console and the Preview tab in Querylab.io’s schema browser read rows without a query job, so they cost nothing.
• Use TABLESAMPLE while you are developing. TABLESAMPLE SYSTEM (1 PERCENT) reads about a hundredth of the data. See TABLESAMPLE.
• Dry-run before you run. The dry run tells you the byte count for free, so a mistake costs nothing.

Related quota errors

Over the wire the message comes back inside an HTTP 403 whose body reads Quota exceeded: ... with "reason": "quotaExceeded", so you may see it with that prefix.

The reason field in the error payload tells you which kind you have. rateLimitExceeded is worth retrying after a few seconds with backoff. quotaExceeded is not. Google’s general advice for it is to wait ten minutes, but that does nothing for this particular quota, which resets monthly.

In Querylab.io

Querylab.io shows the dry-run byte estimate before you run anything, and you can set a maximum GB per query so a query above it does not get sent. See Cost limits and Controlling costs.