Orders per second limits nobody mentions at sign-up
Plenty of brokers cap the number of orders per second, per minute, or per day. The limits live in the fine print, if anywhere, and a scalper running several instruments or a basket-style approach hits them without noticing.
The error message on a breach is rarely helpful. "Too many requests" or a generic failure code shows up in the log. Some platforms retry automatically and the retry itself counts toward the limit, so a minor burst turns into a blocked connection.
Where it hurts: closing a basket of positions at once. Ten close orders fired in the same instant can exceed a limit that opening them one by one never touched. The result is half the basket closed and half still open into a fast move.
Sensible approach is to find out the limits in writing, then pace requests in code with small delays. A hundred milliseconds between orders is usually plenty and costs nothing.
For discretionary traders the same limit applies to rapid-fire clicking. Spamming the button on a requote can trigger the throttle and lock the session for a short while.
Does the broker publish a numeric limit on orders per second, or was it only discovered by hitting it?
Orders per second limits nobody mentions at sign-up
Orders per second limits nobody mentions at sign-up
It’s Fairman 
Re: Orders per second limits nobody mentions at sign-up
Hi Fairman,Fairman wrote: Fri Oct 30, 2026 12:36 pm Orders per second limits nobody mentions at sign-up
Plenty of brokers cap the number of orders per second, per minute, or per day. The limits live in the fine print, if anywhere, and a scalper running several instruments or a basket-style approach hits them without noticing.
The error message on a breach is rarely helpful. "Too many requests" or a generic failure code shows up in the log. Some platforms retry automatically and the retry itself counts toward the limit, so a minor burst turns into a blocked connection.
Where it hurts: closing a basket of positions at once. Ten close orders fired in the same instant can exceed a limit that opening them one by one never touched. The result is half the basket closed and half still open into a fast move.
Sensible approach is to find out the limits in writing, then pace requests in code with small delays. A hundred milliseconds between orders is usually plenty and costs nothing.
For discretionary traders the same limit applies to rapid-fire clicking. Spamming the button on a requote can trigger the throttle and lock the session for a short while.
Does the broker publish a numeric limit on orders per second, or was it only discovered by hitting it?
Most direct-access brokers explicitly publish their rate limits in their API documentation, but retail brokers using off-the-shelf platforms typically hide these thresholds in their server-side configurations, leaving developers to discover them through log errors.
Here is how the industry handles order rate limits and how to structure your execution code to avoid triggering them.
Published vs. Hidden Limits
Published (Direct API Brokers): Institutional and direct-access brokers explicitly define their quotas. For example, Interactive Brokers publishes a strict pacing limit of 10 to 50 requests per second depending on the endpoint. Tradovate restricts users to roughly 80 requests per minute. When these limits are breached, the API reliably returns an HTTP 429 Too Many Requests status code.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Orders per second limits nobody mentions at sign-up
Hidden (Retail Platforms): On retail platforms like MetaTrader 4, MetaTrader 5, and cTrader, these caps are rarely documented at sign-up. Brokers set these limits internally at the server level (e.g., MT5's built-in anti-DDoS and order frequency thresholds) to protect their infrastructure. Traders typically only discover them when their EA or cBot attempts to flatten a heavy pyramid of scaled positions and the server kicks back a "Trade context is busy", "Too many requests", or a generic order rejection error.
The Exit Burst Problem
As you noted, scaling into a position rarely triggers the throttle because the entries are naturally paced. The limit is almost always breached on the exit. Closing 10 or 20 heavily scaled legs simultaneously forces a sudden burst of requests. Furthermore, if a 429 error or server rejection triggers an immediate, unpaced automatic retry loop in the algorithm, the retry itself acts as a multiplier, completely locking the session or resulting in a temporary IP ban.
The Exit Burst Problem
As you noted, scaling into a position rarely triggers the throttle because the entries are naturally paced. The limit is almost always breached on the exit. Closing 10 or 20 heavily scaled legs simultaneously forces a sudden burst of requests. Furthermore, if a 429 error or server rejection triggers an immediate, unpaced automatic retry loop in the algorithm, the retry itself acts as a multiplier, completely locking the session or resulting in a temporary IP ban.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Orders per second limits nobody mentions at sign-up
Architecting the Fix
To safely close a basket of positions across environments like MQL or C#, the execution logic must pace the outbound requests rather than relying on the broker's queue:
Pacing the Loop: The simplest and most effective fix is injecting a short delay. In MQL4/MQL5, adding a 100–200ms Sleep() command at the end of each iteration of your for loop when closing a basket ensures the broker's server has time to process the prior request.
Asynchronous Routing: In cTrader (cAlgo) or custom C# API integrations, using asynchronous methods like ClosePositionAsync paired with an await Task.Delay(100) prevents thread blocking while respecting the broker's limit. In MQL5, using OrderSendAsync pushes the order to the terminal's background thread, but you must still meter the rate at which you feed that function to avoid overwhelming the trade server.
Token Buckets: If you are running multiple EAs across different charts or managing multiple accounts, a simple loop delay isn't enough because the EAs might fire simultaneously. Implementing a token bucket or leaky bucket pattern locally ensures the global outbound request rate across all your algorithms stays below the broker's internal ceiling.
To safely close a basket of positions across environments like MQL or C#, the execution logic must pace the outbound requests rather than relying on the broker's queue:
Pacing the Loop: The simplest and most effective fix is injecting a short delay. In MQL4/MQL5, adding a 100–200ms Sleep() command at the end of each iteration of your for loop when closing a basket ensures the broker's server has time to process the prior request.
Asynchronous Routing: In cTrader (cAlgo) or custom C# API integrations, using asynchronous methods like ClosePositionAsync paired with an await Task.Delay(100) prevents thread blocking while respecting the broker's limit. In MQL5, using OrderSendAsync pushes the order to the terminal's background thread, but you must still meter the rate at which you feed that function to avoid overwhelming the trade server.
Token Buckets: If you are running multiple EAs across different charts or managing multiple accounts, a simple loop delay isn't enough because the EAs might fire simultaneously. Implementing a token bucket or leaky bucket pattern locally ensures the global outbound request rate across all your algorithms stays below the broker's internal ceiling.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Orders per second limits nobody mentions at sign-up
Good write-up, and the token bucket point is the one most people skip. I learned it the hard way with three EAs on one MT5 terminal. Each closed its own basket of 6 legs with a 100ms pause, which looked safe on paper. When a USD news spike hit all three charts at once, that was 18 close requests inside roughly 600ms, about 30 per second, and the server started rejecting somewhere around the tenth. Two legs stayed open for another 8 pips against me at 0.5 lots each, so about 80 dollars for a bug that costs nothing to fix.PTScalper wrote:Implementing a token bucket or leaky bucket pattern locally ensures the global outbound request rate across all your algorithms stays below the broker's internal ceiling.
What I run now: one shared counter through a terminal global variable, capped at 5 orders per second across every EA, and on any rejection a back-off of 500ms, then 1s, then 2s, never an instant retry. Also worth pushing support for the number in writing. Mine eventually admitted 10 per second.
It’s Fairman 
Re: Orders per second limits nobody mentions at sign-up
Fairman, one shared counter across EAs is the only part that actually matters there... pacing each basket on its own was never going to account for correlated news-spike closes. 30 req/sec from three "safe" EAs is exactly the kind of maths that doesnt show up until it does.Fairman wrote:What I run now: one shared counter through a terminal global variable, capped at 5 orders per second across every EA, and on any rejection a back-off of 500ms, then 1s, then 2s, never an instant retry.
Getting the 10/sec in writing is worth more than the fix itself, mind you, since now you can hold them to it if it ever gets revised down without notice.
Re: Orders per second limits nobody mentions at sign-up
That's a good point about the counter being terminal-wide rather than per EA; I learned that the hard way because each EA looked safe in isolation. Keeping the ceiling at 5 per second against a broker-confirmed 10 per second gives some headroom, and the stepped 500 ms, 1 s, then 2 s back-off stops a rejection burst from generating even more requests.
I've also started giving exits priority in the queue, so new entries or amendments can't consume the slots needed to flatten a basket during a spike. I'm logging request timestamps and rejects across the whole terminal in both calm and news conditions. The published figure is useful, but I still want the broker to confirm whether it's per account, terminal, or connection, and whether it changes during volatility. If the limit is connection-level, separate terminals need a shared counter or a single execution gateway; otherwise each one can look safe on its own while the account-wide total breaches it.
I've also started giving exits priority in the queue, so new entries or amendments can't consume the slots needed to flatten a basket during a spike. I'm logging request timestamps and rejects across the whole terminal in both calm and news conditions. The published figure is useful, but I still want the broker to confirm whether it's per account, terminal, or connection, and whether it changes during volatility. If the limit is connection-level, separate terminals need a shared counter or a single execution gateway; otherwise each one can look safe on its own while the account-wide total breaches it.
It’s Fairman 