Node limit
To pass schema validation, all GraphQL API v4 calls must meet these standards:
- Clients must supply a
first
orlast
argument on any connection. - Values of
first
andlast
must be within 1-100. - Individual calls cannot request more than 500,000 total nodes.
Calculating nodes in a call
These two examples show how to calculate the total nodes in a call.
-
Simple query:
query { viewer { repositories(first: 50) { edges { repository:node { name
issues(first: <span class="greenbox">10</span>) { totalCount edges { node { title bodyHTML } } } } } }
} }
Calculation:
50 = 50 repositories
-
50 x 10 = 500 repository issues
= 550 total nodes</pre>
-
-
Complex query:
query { viewer { repositories(first: 50) { edges { repository:node { name
pullRequests(first: <span class="greenbox">20</span>) { edges { pullRequest:node { title comments(first: <span class="bluebox">10</span>) { edges { comment:node { bodyHTML } } } } } } issues(first: <span class="greenbox">20</span>) { totalCount edges { issue:node { title bodyHTML comments(first: <span class="bluebox">10</span>) { edges { comment:node { bodyHTML } } } } } } } } } followers(first: <span class="bluebox">10</span>) { edges { follower:node { login } } }
} }
Calculation:
50 = 50 repositories
-
50 x 20 = 1,000 pullRequests
-
50 x 20 x 10 = 10,000 pullRequest comments
-
50 x 20 = 1,000 issues
-
50 x 20 x 10 = 10,000 issue comments
-
10 = 10 followers
= 22,060 total nodes</pre>
-
Rate limit
The GraphQL API v4 limit is different from the REST API v3's rate limits.
Why are the API rate limits different? With GraphQL, one GraphQL call can replace multiple REST calls. A single complex GraphQL call could be the equivalent of thousands of REST requests. While a single GraphQL call would fall well below the REST API rate limit, the query might be just as expensive for GitHub's servers to compute.
To accurately represent the server cost of a query, the GraphQL API v4 calculates a call's rate limit score based on a normalized scale of points. A query's score factors in first and last arguments on a parent connection and its children.
- The formula uses the
first
andlast
arguments on a parent connection and its children to pre-calculate the potential load on GitHub's systems, such as MySQL, ElasticSearch, and Git. - Each new connection has its own point value. Points are combined with other points from the call into an overall rate limit score.
The GraphQL API v4 rate limit is 5,000 points per hour.
Note that 5,000 points per hour is not the same as 5,000 calls per hour: the GraphQL API v4 and REST API v3 use different rate limits.
Note: The current formula and rate limit are subject to change as we observe how developers use the GraphQL API v4.
Returning a call's rate limit status
With the REST API v3, you can check the rate limit status by inspecting the returned HTTP headers.
With the GraphQL API v4, you can check the rate limit status by querying fields on the rateLimit
object:
query {
viewer {
login
}
rateLimit {
limit
cost
remaining
resetAt
}
}
-
The
limit
field returns the maximum number of points the client is permitted to consume in a 60-minute window. -
The
cost
field returns the point cost for the current call that counts against the rate limit. -
The
remaining
field returns the number of points remaining in the current rate limit window.) -
The
resetAt
field returns the time at which the current rate limit window resets in UTC epoch seconds.
Calculating a rate limit score before running the call
Querying the rateLimit
object returns a call's score, but running the call counts against the limit. To avoid this dilemma, you can calculate the score of a call before you run it. The following calculation works out to roughly the same cost that rateLimit { cost }
returns.
- Add up the number of requests needed to fulfill each unique connection in the call. Assume every request will reach the
first
orlast
argument limits. - Divide the number by 100 and round the result to get the final aggregate cost. This step normalizes large numbers.
Note: The minimum cost of a call to the GraphQL API v4 is 1, representing a single request.
Here's an example query and score calculation:
query {
viewer {
login
repositories(first: 100) {
edges {
node {
id
issues(first: 50) {
edges {
node {
id
labels(first: 60) {
edges {
node {
id
name
}
}
}
}
}
}
}
}
}
}
}
This query requires 5,101 requests to fulfill:
- Although we're returning 100 repositories, the API has to connect to the viewer's account once to get the list of repositories. So, requests for repositories = 1
- Although we're returning 50 issues, the API has to connect to each of the 100 repositories to get the list of issues. So, requests for issues = 100
- Although we're returning 60 labels, the API has to connect to each of the 5,000 potential total issues to get the list of labels. So, requests for labels = 5,000
- Total = 5,101
Dividing by 100 and rounding gives us the final score of the query: 51