is anyone using Wasm UDFs in production? We are currently looking into it for some processing tasks. (some pretty basic text processing: normalising text, extracting entities e.g. hashtags from text and so on)
We somehow get some totally inconsistent behaviour: random errors (sometimes the query works fine, sometimes it throws), different output for the same input locally (using writ-docker) and on the db. ‘SHOW WASM BACKTRACE;’ always returns no results, even if it is called immediately after an error. So there is basically no way to debug the errors… Or am I missing something? Is this feature still in an experimental state?
I was able to resolve the issue: It seems like that a WASM UDF is cached for the current query shape once it’s executed the first time & this cache doesn’t get purged if a new version of the function is uploaded. So the issue was basically a broken cached function. I just had to run each “try” with another query shape or clear the cache Is there anyway to submit a bug report to the dev team? Also we are currently implementing a lot of stuff in rust to move some currently pre-computed analytical tasks to a query-time execution model. Mostly natural language processing. In case we can help to further improve the DX and performance of SingleStore WASM integration: feel free to reach out to me happy to provide some feedback/test cases
Hi @tom – yes, Wasm is a supported feature and customers are using it in production. In fact, we use it a lot in Kai, our Mongo-compatible API implementation. Were you able to file a bug report? What release version are you using?
If you are a paying customer, and you have not filed the bug report yet, the best way to file it is to open a support ticket and provide the repro details.