I'm looking at LlamaHub loaders for a project. Need to fetch data from several sources at once.
Is there a limit to parallel requests? Like, if I fire off 10 loader instances, will they throttle or fail? Don't want to hit a hidden rate limit and crash things. Free tier concerns mostly.
The loader's parallel request limit isn't what you need to worry about, it's the underlying API you're fetching from. Each source has its own throttle.
Check their documentation for rate limits on their free tier - I've seen them range from 10 to 60 requests per minute. Start with a small batch and implement some basic error handling for 429 codes, maybe with exponential backoff.
If you're using the same API credential across all 10 instances, you'll hit that ceiling fast.
get it in writing
Exactly. "Each source has its own throttle" is the key. I found this out the hard way.
Some SaaS tools also have hidden concurrent connection limits on their side, separate from the requests-per-minute. You can get queued or dropped before you even hit a 429.
What's a good way to test for that besides just watching for failures?
That's a good point about hidden connection limits. For testing, I've had some luck with a staggered start - like launching instances 5 seconds apart and logging connection times. If the later ones consistently take longer to establish, that's a clue about queuing.
Is there a way to actually see those limits in something like Chrome DevTools during a normal browser session, before writing any loader code? Might be easier to spot.
One step at a time