Timeouts with more than 20 seconds

Hello everyone,
We found out that some requests to Vectara on August 14th timed out.
We are using gRPC query method.
Can you help us to find out what happened on your side?
If you need any further details like account number, please let me know.

The timestamps are GMT+3.

Hi, that time period looks normal for us in our logs. Can you tell us what account and corpus this is? Additionally, queries with summaries can take longer than normal occasionally.

Here is the account number: 1353344576
Corpus id: 24

I do some slow queries for you around that time period, but not that long. But no query was that long in internal logs, and not especially near your indicated times. It’s possible some network device between us and you had transient issues at the time. Is this an issue you are still seeing?

Looks like you are using hybrid (keyword) search with lambda value > 0. Can you please confirm that?

Yes we are using hybrid search with lambda value = 0.075

We have switch to a new corpus this week (corpus id: 31) and we have logged 5 timeouts to Vectara in just 2 days

2023-09-03 21:41:28,432 Deadline Exceeded
2023-09-03 21:41:28,432 Query: coral
2023-09-03 21:41:28,432 Request duration: 10.209462642669678 seconds
2023-09-03 22:10:21,011 Deadline Exceeded
2023-09-03 22:10:21,011 Query: qur'an rubies
2023-09-03 22:10:21,012 Request duration: 10.251399755477905 seconds
2023-09-03 22:14:25,451 Deadline Exceeded
2023-09-03 22:14:25,451 Query: ibn kathir last al baqarah
2023-09-03 22:14:25,452 Request duration: 10.202023267745972 seconds
2023-09-03 22:25:02,331 Deadline Exceeded
2023-09-03 22:25:02,332 Query: allah forgive mistakes
2023-09-03 22:25:02,332 Request duration: 10.171724557876587 seconds
2023-09-04 03:52:27,254 Deadline Exceeded
2023-09-04 03:52:27,254 Query: surat albaqoroh cz the reward of having a family
2023-09-04 03:52:27,255 Request duration: 10.167435646057129 seconds

We have changed the lambda value to 0.35 and we are using the new encoder Boomerang_230822.

Sorry for the late response. Yes, we are aware of the high latency when lambda > 0, we are currently working on it. ETA is 3 weeks.