Why Is My Supabase Egress So High? How to Find the Cause
Trace high Supabase egress to the project, service and workload driving it. Learn how to tell normal growth from a wasteful query or access issue.
Nov 22, 2026 · 4 min read
Field Notes from Thunkle, a studio that takes AI-built apps from prototype to secure, production-ready software.
Supabase egress is data sent out from Supabase. A rise can come from ordinary customer activity, an oversized API response, repeated file downloads or a background job doing more work than intended. Start with the project, service and time of the spike. High egress alone does not tell you whether a bot is responsible or whether anything private was exposed.
1. Find the project and service driving the increase
Open Usage for the organisation. Narrow the time range to the spike, then select projects or preview branches individually. The organisation total can include a branch that doesn't appear in its parent project's own figure. Check the service breakdown and the separate cached egress series. A bigger organisation total isn't necessarily a problem in your main app. Supabase's egress guide shows where these numbers come from.
Record the first day usage changed, the affected project and service, and what shipped or ran around that time. Compare a similar period before the change. If you sell more on weekends, don't treat a weekend-to-weekday difference as evidence of a bug. Check usage against the plan and invoice separately; a rise in data transfer and an overage charge aren't the same observation.
2. Work out whether you sent more responses or bigger ones
If the Database or Data API is driving the spike, look for heavily requested paths in Logs Explorer and frequent queries in Query Performance. Check how often the relevant query runs and how many rows it returns. Supabase notes that the API Gateway log view doesn't include response bytes, so the busiest path is a lead, not proof that it used the most bandwidth. Check a representative response on your own app with test data, or measure payload sizes in application telemetry without recording the payload itself.
Example, not a customer incident: a dashboard makes one 2 MB request on every refresh. If it refreshes 100 times during a busy hour, that is roughly 200 MB of response data before accounting for compression and other differences. Cutting the response to the fields the dashboard renders can help more than trimming a few calls. If the payload is small but the same request runs every few seconds in an idle tab, change the polling instead. Measure the same workflow again before declaring it fixed.
If Storage is driving it, check which files the app serves and when. Common explanations include an original image loaded for every thumbnail, a large file fetched more than once during a screen transition, or expected use after launching a feature. Use smaller assets when they serve the same purpose and keep caching consistent with the file's access rules. Don't make a private bucket public to save bandwidth. See Supabase's Storage bandwidth guide.
If the rise tracks a worker, export or sync, compare usage with its run times and item counts. Does it fetch the full dataset every run, or only the changes it needs? Did a retry or overlapping schedule cause the same batch to run twice? Keep the backup or business process you need, but check that its actual frequency and volume match the plan.
3. Decide whether it's a usage problem or an access problem
Start with evidence. A usage spike that follows a product launch and scales with known downloads may be healthy. A sharp rise on one route with no corresponding product activity warrants a closer look, but a user agent or IP address alone won't tell you who is responsible. Compare the spike with deployment changes, traffic and job activity before blaming bots.
Treat unexpected access to private data separately from the bill. If you see evidence that private records or files may have been returned to someone who should not have access, preserve relevant logs and get the project reviewed promptly. Ask the owner to check the affected permissions and credentials. Redact tokens, customer data and signed file links when sharing evidence. A browser-visible publishable key by itself doesn't prove a leak; what matters is what callers can read or change. See which Supabase keys belong in the frontend.
4. Fix the biggest contributor and check it again
- Too much data per response: select only the fields and rows the screen needs, page long lists, and avoid returning an entire record after an update when the client doesn't use it.
- Too many calls: check duplicate requests, polling and overlapping jobs. Cache only where the response is appropriate for the intended audience and freshness requirements.
- Large or repeated file downloads: use smaller assets and avoid refetching the same file without a product reason. Confirm private files stay private.
- Unexplained access: have a developer review the affected Data API, storage rules and callers, using approved access and an isolated test environment for verification.
Compare both data transferred per workflow and overall service usage after the fix. A smaller response on the test screen is useful; it doesn't prove the organisation-wide spike has stopped. Keep the timing, affected path or job, change made and before-and-after evidence together.
If you can't account for the spike, Thunkle's Supabase audit can trace the application calls, permission boundaries and backend jobs behind it. Bring the date range, affected project, service breakdown and a redacted log sample. Don't send service-role keys or customer data. If you already know the cause and need the fix implemented, see SaaS development and integrations.
Find out why your Supabase usage changed.
Share the affected project, the time of the spike and the service driving it. Thunkle can scope a review of the queries, jobs and permissions behind the usage.
