Exchange Rate Limits Overview
Summary
Private endpoints are authenticated.
REST API Rate Limits
Public Endpoints
- Requests per second per IP: VAR::REST_PUB_REQS_PER_SEC_PER_IP
- Requests per second per IP in bursts: Up to VAR::REST_PUB_REQS_PER_SEC_PER_IP_BURST
Private Endpoints
- Requests per second per profile: VAR::REST_PRV_REQS_PER_SEC_PER_PROFILE
- Requests per second per profile in bursts: Up to VAR::REST_PRV_REQS_PER_SEC_PER_PROFILE_BURST
Private /fills Endpoint
- Requests per second per profile: VAR::REST_PRV_FILLS_REQS_PER_SEC_PER_PROFILE
- Requests per second per profile in bursts: Up to VAR::REST_PRV_FILLS_REQS_PER_SEC_PER_PROFILE_BURST
Private /loans Endpoint
- Requests per second per profile: VAR::REST_PRV_LOANS_REQS_PER_SEC_PER_PROFILE
Rate limits do not apply to List loan assets (/loans/assets) which is not private.
FIX API Rate Limits
FIX 4.2 Rate Limits
- Requests per rolling second per session: VAR::FIX_REQS_PER_ROLLING_SEC_PER_SESSION
- Messages per second in bursts: VAR::FIX_MSGS_PER_SEC_BURST
FIX 5.0 Rate Limits
- VAR::FIX_FIVE_LOGON_PER_SEC logons per second per API key
- VAR::FIX_FIVE_REQS_PER_SEC requests per second
Your FIX 5 session is disconnected if your messages exceed VAR::FIX_FIVE_REQS_PER_SEC_DISCONNECT messages per second
FIX Maximums
- Maximum API keys per session/connection: VAR::FIX_MAX_API_KEYS_PER_SESSION
- Maximum connections per profile: VAR::FIX_MAX_CONNECTIONS_PER_PROFILE . See FIX Best Practices.
- Maximum connections per user across all profiles: VAR::FIX_MAX_CONNECTIONS_PER_USER
- Maximum profiles per user: VAR::FIX_MAX_PROFILES_PER_USER
- Maximum orders per batch message message (new and cancelled): VAR::FIX_MAX_BATCH_ORDERS
Websocket Rate Limits
- Requests per second per IP: VAR::WS_REQS_PER_SEC_PER_IP
- Requests per second per IP in bursts: Up to VAR::WS_REQS_PER_SEC_PER_IP_BURST
- Messages sent by the client every second per IP: VAR::WS_CLIENT_MSGS_PER_SEC_PER_IP
Other
- Maximum open orders: VAR::MAX_OPEN_ORDERS
How Rate Limits Work
Rate-limiting for both the Exchange REST API and the FIX API use a lazy-fill token bucket implementation.
A TokenBucket stores a maximum amount of tokens, which is the burst size, and fills at a given rate called the refresh rate. The bucket starts full, and as requests are received, a token is removed for each request. Tokens are continuously added to the bucket at the refresh rate until full.
When a user sends a request, the TokenBucket calculates whether or not to rate limit the user as follows:
- Fill the user’s TokenBucket to a token size based on the following formula:
token_amount = min(burst, previous_token_amount + (current_time - previous_request_time) * refresh_rate) - Remove 1 token if possible, otherwise rate limit the request.
- Repeat Steps 1 and 2 for each subsequent request.
TokenBucket Example
Let’s say you have a TokenBucket with burst = 3 and refresh_rate = 1. The table below represents the state of your token bucket after a series of requests:
| Action | Time | Tokens | Notes |
|---|---|---|---|
| Initial State | 0.0 | 3.0 | New TokenBucket is initialized to max capacity (burst) |
| Request 1 | 0.5 | 2.0 | Fill TokenBucket, then remove a token, because we are at max capacity, and subtract 1 token from 3 |
| Request 2 | 0.8 | 1.3 | Fill TokenBucket to 2.3 (min(3, (2 + (.8 - .5) * 1.0)) = min(3, 2.3) = 2.3), then subtract 1 |
| Request 3 | 0.9 | 0.4 | Fill TokenBucket to 1.4 (min(3, (1.3 + (.9 - .8) * 1.0)) = min(3, 1.4) = 1.4), then subtract 1 |
| Request 4 | 1.0 | 0.5 | Fill TokenBucket to 0.5 (min(3, (.4 + (1.0 - .9) * 1.0)) = min(3, 0.5) = 0.5). Ratelimit because we don’t have enough tokens available |
| Request 5 | 1.4 | 0.9 | Fill TokenBucket to 0.9 (min(3, (0.5 + (1.4 - 1.0) * 1.0)) = min(3, 0.9) = 0.9). Ratelimit because we don’t have enough tokens available |
| Request 6 | 1.8 | 0.3 | Fill TokenBucket to 1.3 (min(3, (0.9 + (1.8 - 1.4) * 1.0)) = min(3, 1.3) = 1.3), then remove 1 |
| Request 7 | 5.0 | 2.0 | Fill TokenBucket to 3.0 (min(3, (0.3 + (5.0 - 1.8) * 1.0)) = min(3, 3.5) = 3), since we would “overflow” with our calculations, then subtract 1 |