DEV Community

Fabian
Fabian

Posted on

Apify limits I only found out about by hitting them

I've been publishing scrapers on the Apify Store since August 2026, 24 public actors at this point. Most of the platform is well documented, but a handful of limits I only learned about when a request came back with an error code I hadn't seen before. Here they are, with the error you get, so the next person can search for it.

Publishing and the Store

5 actor publications per day. Making an actor public (or re-publishing it) counts against a daily budget of 5. The sixth one returns daily-publication-limit-exceeded. Hand-retrying in a loop doesn't help, it resets daily. I ended up with a small timer script that retries the pending ones once in a while.

Example tasks: 50 per actor in the docs, 10 published per actor in practice. Published tasks get their own landing page under apify.com/<user>/<actor>/examples/<task>, which is great for search. The docs say 50 per actor. What I ran into was too-many-public-actor-tasks-per-actor at 10, plus an account-wide cap of 100 published tasks. So with more than 10 actors you have to decide which ones get pages.

Publishing needs an output schema now. PUT /v2/acts/{id} with isPublic: true fails with schemas-required if the default build has no output section in .actor/actor.json. Adding it means a rebuild first, then the publish works.

Icons can't be set through the API. pictureUrl only accepts Apify's own uploads, so icons are a manual step in the Console. Reading the field back through GET /v2/acts/{id} at least tells you which actors still have none.

defaultRunOptions in actor.json was not applied by apify push. I set default memory in the file and the actor kept the old value. Setting it through the API (PUT /v2/acts/{id} with defaultRunOptions) worked.

Pricing

Price increases start 14 days in the future. Pay-per-event prices are a list of pricingInfos entries. A new entry that raises a price has to start at least 14 days out, and Apify then emails existing users. Decreases can take effect immediately.

One price change per actor per month. Any new pricingInfos entry counts, including a decrease or a fix to an increase that hasn't started yet. The second one in a month returns HTTP 403 cannot-modify-actor-pricing-too-frequently, with the next allowed date. A scheduled entry also can't be edited or deleted: the PUT has to contain the existing entries unchanged plus at most one new one. So measure first, then set the price.

The start event is charged per GB. apify-actor-start is charged automatically by the platform (calling Actor.charge on it fails). It's charged once for the first GB of memory and once more for every additional GB. A $0.01 start price on an actor with 4 GB default memory is $0.04 per run. My HTTP scrapers peak at 50 to 130 MB, so 1024 MB is now the default everywhere.

Running things

Concurrent runs: HTTP 402, and no run at all. Testing my own paid actors, starting about 5 runs at once returned HTTP 402 on run creation. No run is created, so nothing shows up as failed anywhere. I test in batches of 3 now.

The plan's usage limit is a hard stop. On the free plan the $5 monthly credit is a cap: when it's used up, runs and API calls are blocked until the usage cycle resets. The account also dropped back to the free plan once without me noticing, so before a test batch I read GET /v2/users/me/limits instead of assuming the plan.

Maximum cost per run ends as ABORTED. When a pay-per-event run reaches the user's max cost, the platform aborts it, often before the actor can finish cleanly, and an aborted run counts as a failure. The fix is to cap your own pushes with the SDK's calculateMaxEventChargeCountWithinLimit, keep one event price in reserve, and exit as SUCCEEDED with a status message that says the budget was reached. The minimum maxTotalChargeUsd a user can start with is $0.05.

Debugging other people's runs

You can't see them. As the owner of a public actor you get no run object, no log and no input for runs started by other users. GET /v2/acts/{id}/runs returns only your own. What you do get is stats.publicActorRunStats30Days, a 30-day count of runs per status. I tried to get around this with webhooks; that turned out to have the same limit, which is its own post.

The practical consequence: the status message is the only thing the user sees when a run fails, and the 30-day failed count is the only thing you see. Both are worth getting right.

Required fields with a default let empty runs start. The platform validates input against the schema before starting the container. A missing required field returns HTTP 400 and no run exists. But a required field that also has "default": [] passes that check, the container starts, your code throws, and the run counts as FAILED in public stats. Required fields get a prefill and minItems: 1, never a default.

None of this is hidden, exactly. It's spread across docs, error messages and the API reference. If you're about to publish your first actor, the 402 on concurrent runs and the one-price-change-per-month rule are the two I'd want to know on day one.

Top comments (0)