Ed Welch

hey @marcusteixeira thanks so much for the detailed info here, sorry I didn't get back to this sooner. Really appreciate you trying out various things. I think we need to...

we were talking today about logging what the WAL is doing while it replays, I'm suspicious that maybe when there is a really big wal to replay it can take...

We use snappy everywhere, gzip compresses more than snappy but is much more cpu expensive at query time so we use snappy for the improved performance. how many write nodes...

are you using a replication factor > 1? If so, this would multiply the amount of data each ingester processes by the replication_factor. My sort of gut feeling here is...

I think almost every instance in this ticket relates to having insufficient network bandwidth for flushing chunks to storage, or sending too much data into too few ingesters. @afayngelerindbx has...

@afayngelerindbx >Looking at the code, it would be fairly simple to limit the number of chunks per flush to some configurable threshold. This, I believe, would make it less likely...

Ok, have an update here and @DylanGuedes and I took a look at this and he is gonna work on a change, Summary: the current code doesn't take credit for...

Also thanks so much @afayngelerindbx for digging on this, your sleuthing has been a big help, and all the graphs and due diligence by everyone else has been super helpful...

I know this will be a bit tedious but can we generate a list of all the metric names that will be changed? We should include this list in the...

`fake` is the name of the tenant when auth is not enabled, when auth is enabled the folder name would be whatever the tenant ID is. It's an unfortunate naming...