WAL in Ingesters (chunks based)
The ingesters used to lose data whenever they would crash as it held all the data in the memory. I took over Tom's initial PoC and got Write-Ahead-Log (WAL) into ingesters in PR#1103 which prevents data loss during crashes. Read more about it in this blog post.
In the process of massively scaling alertmanagers in Cortex, I implemented the distributor component for alertmanager in PR#3671 which shards and replicates the requests across alertmanagers.