Asdine El Hrychy

It is a bit difficult to create a `consumer.Message` or `producer.Message` the first time when testing our handlers. It would be nice if we could have some kind of helper...

Replace logger by logrus or equivalent

2

In order to better control the behaviour of the logs and be able for example to send everything that was logged on Error level to Sentry, we need a library...

When a handler panics we need to recover and retry to prevent the service to crash. - recover - log on Error level - retry after the retry delay

various fixes to indexing queue

1

### What's being changed: This PR fixes a few issues with the indexing queue and adds two env vars: - `ASYNC_MAX_CHUNKS_PER_TICK`: to control the dequeue speed, i.e how many batches...

do not merge: temporary branch for list leak investigation

1

### What's being changed: ### Review checklist - [ ] Documentation has been updated, if necessary. Link to changed documentation: - [ ] Chaos pipeline run or not necessary. Link...

hnsw: set default max connections to 32

1

### What's being changed: Fixes #4363  - orange: 64 - blue: 32 ### Review checklist - [ ] Documentation has been updated, if necessary. Link to changed documentation: -...

The drift files are a very convenient way of maintaining SQL queries but because the syntax is not exactly SQL, the IDE formatters (VS Code, Goland, etc.) can't work properly...

on-disk async indexing

1

### What's being changed: ### Review checklist - [ ] Documentation has been updated, if necessary. Link to changed documentation: - [ ] Chaos pipeline run or not necessary. Link...

Basically, each queue updates the status of the shard depending on its size (= 0 -> shard ready, > 0 shard indexing), but since target vectors have each their own...

async deletions

1

### What's being changed: Deletions are now asynchronous when async indexing is enabled. The queue treats them as a regular operation and dequeues inserts and deletions in order. However, when...