Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore

#1

Post by bauerpl » Fri, Jul 27, 2018, 7:24 AM UTC

Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore
eng.uber.com

Re: Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore

#2

Post by damm » Fri, Jul 27, 2018, 7:33 AM UTC

Go ahead and downvote this; but where's the source?

Otherwise isn't this pointless bragging?

Re: Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore

#3

Post by miohtama » Sat, Jul 28, 2018, 3:22 AM UTC

damm wrote: post #2

Go ahead and downvote this; but where's the source? Otherwise isn't this pointless bragging?

It is not pointless for Uber. Please read it as a job advertisement for scalable database experts.

Re: Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore

#4

Post by takeda » Sat, Jul 28, 2018, 3:46 AM UTC

damm wrote: post #2

Go ahead and downvote this; but where's the source? Otherwise isn't this pointless bragging?

So according to this post they created geographically distributed, multi master (with a mesh topology) that is always consistent.

Either they created a holy grail of databases, or an opportunity to write a great Post-Mortem in the future.

Re: Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore

#5

Post by thdxr » Sat, Jul 28, 2018, 4:54 AM UTC

takeda wrote: post #4

damm wrote: post #2

Go ahead and downvote this; but where's the source? Otherwise isn't this pointless bragging?

So according to this post they created geographically distributed, multi master (with a mesh topology) that is always consistent. Either they created a holy grail of databases, or an opportunity to write a great Post-Mortem in the future.

Or it's just incredibly slow and not always available

Re: Herb: Multi-DC Replication Engine for Uber’s Schemaless Datastore

#6

Post by throwaway123402 » Sat, Jul 28, 2018, 5:38 AM UTC

takeda wrote: post #4

damm wrote: post #2

Go ahead and downvote this; but where's the source? Otherwise isn't this pointless bragging?

So according to this post they created geographically distributed, multi master (with a mesh topology) that is always consistent. Either they created a holy grail of databases, or an opportunity to write a great Post-Mortem in the future.

It's a little unclear from the article, but based on

> Since the system supports asynchronous replication, our data centers eventually become consistent. Writes are considered complete as soon as the remote data center acknowledges it.

I assume its eventually consistent (so AP).

I wish they compared why normal mysql async replication fails for them and they needed to build this though.