Writing

Search Console URL Inspection API quota: why 429 Quota exceeded outlasts midnight

By Mohamed Gado · · 4 min read

The Search Console URL Inspection API allows 2,000 inspections a day and 600 a minute per property. Don't schedule a job around the moment you think that daily allowance resets. My daily index report inspects all 790 URLs in my sitemap, and on 19 November one inspection at 11:24 UTC still returned 429 Quota exceeded, more than three hours after a run that I had called safe. The fix was to stop guessing: the job now spends one inspection as a probe, and it retries every hour until the probe passes.

What are the URL Inspection API limits?#

Google documents two limits on urlInspection.index.inspect, both per Search Console property:

Limit Value
Queries per day 2,000
Queries per minute 600

The rest of the Search Console API (search analytics, sitemaps) has its own, much larger allowances. The inspection quota is the tight one. A site with 790 URLs can inspect every page about twice a day, and only if nothing else uses the quota.

Why does 429 Quota exceeded last longer than you expect?#

I assumed the quota would free up at a fixed time each day. On 19 November I had run the full report twice by hand, at 03:47 and 04:02 UTC: about 1,580 inspections, plus a few local test runs. The next scheduled run was at 08:05 UTC, so I wrote in my notes that it would "have the window to itself".

At 11:24 UTC I made one real inspection, and it returned 429.

I can't see Google's accounting, so I don't know exactly how the window works. What I can say from my own logs is that inspections made at 04:00 UTC were still counted against me more than seven hours later. Midnight Pacific time, when Google's daily API quotas usually reset, was 08:00 UTC that day, between those two moments, and it didn't clear them. In practice the quota behaved like a rolling window. A schedule built on a reset time is a guess, and nothing in my code had tested that guess.

Probe the quota before spending it#

The report now runs in this order:

  1. Done? If today's rows already exist, stop. No API calls.
  2. Probe. Inspect one real URL. A 429 ends the run here, before it writes anything or starts the 790-URL fan-out.
  3. Fan out. Inspect every URL in batches, staggered so each batch stays under 600 a minute.
  4. Mail the report.

A failed probe costs one inspection. Without the probe, a run with an exhausted quota would start 790 inspections, store 790 errors, and mail a report full of them.

Retry every hour, and say so when a day is lost#

A cron trigger starts the report every hour from 08:05 to 20:05 UTC:

// The hourly cron tries the report from 08:05 to 20:05 UTC until one run gets past its quota
// probe; later ticks see today's rows and stop at once. 08:05 is not a reset time: the
// inspection quota frees on a rolling window, and on 19 Nov 2026 it was still spent at 11:24 UTC.
export const REPORT_FIRST_HOUR_UTC = 8
export const REPORT_LAST_HOUR_UTC = 20

The first tick whose probe passes does the whole report. Every later tick hits the "done?" step and stops at once. The 20:05 tick is marked final: if the quota is still spent then, it mails "skipped, GSC quota spent all day". A lost day is reported, never silent.

Count a day as done only when the results are real#

The first version had a second bug. The "done?" check looked for today's rows, and a run that stored 790 rows of 429 errors also counted as done. Now a day counts only when most of its rows say something about the URL itself. Rows that only say the request failed (a 429, a 5xx, a timeout, a platform limit) don't count, and the next hourly tick runs the day again.

That check matches failures by kind, not by one error string. It already paid off the next day, when a different failure (a Workers subrequest limit) filled a day with errors that weren't 429s.

Test the pipeline without spending quota#

Inspections are the scarce resource, so the local test harness fakes only the inspection call. Everything else is real: the OAuth token, the Sitemaps API, my live sitemap.xml, the batching and the mail. One local run covers the whole pipeline at the cost of zero inspections. When I do want to know whether the real quota is free, I make exactly one inspection.

Takeaways#

  • The URL Inspection API allows 2,000 inspections a day and 600 a minute per property. Plan for about two full passes of a 790-URL site per day, at most.
  • Don't assume when the quota resets. Probe with one inspection and retry on a schedule.
  • Stop before the fan-out when the probe fails, and mail an explicit "skipped" on the last try.
  • A day full of 429s is not a finished day. Count only results that describe the URL.

For why a URL that Google knows about still isn't in the index, see Discovered, currently not indexed.