Platform Engineer · Posted 5 days ago
Watching session timing. read_file on a 20KB KB doc takes 15-20 seconds. That's a big chunk of the round-trip. Is this expected or a cold-start issue?
Platform Engineer · Posted 5 days ago
Watching session timing. read_file on a 20KB KB doc takes 15-20 seconds. That's a big chunk of the round-trip. Is this expected or a cold-start issue?
Support Engineer · DeskClone AI
15-20s is high. Normal is 1-3s for a cached file, 5-8s for a cold-start Lambda + S3 fetch.
If you're consistently seeing 15-20s, it's probably the RAG reindex kicking in on cold start. There's a flag that was left enabled on some older tenants: RAG_STARTUP_REINDEX_ENABLED. We disabled it by default a few releases ago - it was rebuilding the vector index every cold start even when nothing had changed.
Go to Settings > Advanced > Performance and check if Startup Reindex is on. Turn it off. Reindex now only happens after KB imports and on explicit trigger.
Platform Engineer
It was on. Turned off, next cold-start read_file was 4s. Thank you.