FlowiseAI
[BUG] Memory leak on API upsert request
Describe the bug When i use API request to upsert document to vector store i get memory leak, For 3k items it usses additional 5GB of ram and it doesn't free up. It fills up memory until PC crashes.
I setup TextFile and enabled API and enabled Override Configuration for text file. I'm sending this file via API curl requests from another system
To Reproduce here is simple flow that leaks
Expected behavior Memory would free up after some time
Setup
- Installation docker,
- Flowise Version latest
- OS: [Linux]
Can you try using the Document Store Upsert API? https://docs.flowiseai.com/using-flowise/document-stores#upsert-api
I already tried with document store, but i really don't need to have additional store for the data. I just need to send the data to vectore store in one go, which does works nicely by sending text, but unfortunately over time it leaks memory.
This is might connected: https://github.com/FlowiseAI/Flowise/discussions/3733 I tried to use @bernejorge image but it just won't start the container.
I already tried with document store, but i really don't need to have additional store for the data. I just need to send the data to vectore store in one go, which does works nicely by sending text, but unfortunately over time it leaks memory.
This is might connected: #3733 I tried to use @bernejorge image but it just won't start the container.
I see a similar memory leak issue with Custom MCPs, losing around 150MB every time you add a Custom MCP node (even deleting all flows will not relase it). Over time, during iterative testing, memory consumption can escalate to around 3 or 4 GB easily.
The only way I found to reclaim the leaked memory was by restarting the Flowise container via the implementation of a cron job on the host server to automate the restart process.
I already tried with document store, but i really don't need to have additional store for the data. I just need to send the data to vectore store in one go, which does works nicely by sending text, but unfortunately over time it leaks memory. This is might connected: #3733 I tried to use @bernejorge image but it just won't start the container.
I see a similar memory leak issue with Custom MCPs, losing around 150MB every time you add a Custom MCP node (even deleting all flows will not relase it). Over time, during iterative testing, memory consumption can escalate to around 3 or 4 GB easily.
The only way I found to reclaim the leaked memory was by restarting the Flowise container via the implementation of a cron job on the host server to automate the restart process.
that was already resolved with the fix https://github.com/FlowiseAI/Flowise/pull/4375 no?
that was already resolved with the fix #4375 no?
No, you leave it open for further investigation https://github.com/FlowiseAI/Flowise/issues/4369#issuecomment-2853106883
Hi, facingg same issue in agent v2, for sequential agentflow, any update ?
no database connecttion is thr in our flow, it is coming in agent v2 sequential workflow , llm calls
I can confirm tha PR #5315 fixes this original issue. But it seems like there are more memory issues. It might be worth to create another issues with flow examples and more info on how to reproduce the issue @AbhishekHupele @toi500 cc: @HenryHengZJ