Qdrant Recall Inconsistency: It Took 300 Test Runs to Discover the Index Wasn't Refreshed
At 2 AM, a user reported that the AI Agent suddenly forgot details of a project we discussed yesterday. I groggily opened Grafana and saw the memory recall rate had dropped from 98% to 60%. My first thought was the embedding model acting up again, but after digging through logs, I found a "time gap" between Qdrant writes and queries—data was upserted, yet queries intermittently returned nothing.…
In 2 AM, a user reported that the AI Agent had forgotten details of a project discussed the previous day. The memory recall rate had dropped from 98% to 60%. Upon investigation, it was discovered that there was a time gap between Qdrant writes and queries, causing intermittent failures in query results. This issue was not isolated, as it had occurred multiple times before.
The reporter decided to automate recall consistency testing using pytest and Qdrant to uncover the deeper problems. The key finding was that Qdrant's write and index building are asynchronous, which means that upsert method returns without waiting for index refresh by default. Consequently, subsequent queries may return empty or partial results.
To address this issue, the reporter added a time delay in the code, but this is not an ideal engineering practice as the sleep duration is unpredictable, and treating the symptom rather than the root cause.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
