27 lines
1.5 KiB
Plaintext
27 lines
1.5 KiB
Plaintext
It would be my pleasure to continue serving as Swift PTL for the Yoga cycle.
|
|
|
|
This past cycle, we have continued our commitment to operational excellence.
|
|
We've improved several features vital to large clusters, such as container
|
|
sharding, partition power increases, and storage-policy reconciliation. We've
|
|
improved the durability of erasure-coded data immediately after a write. We've
|
|
made our logs, metrics, and errors more useful.
|
|
|
|
We focus on operators for several reasons. Operators are at the junction of the
|
|
software we write, the hardware we use, and the clients we serve. They have
|
|
perhaps the best perspective to see the value-to-effort ratio of prospective
|
|
improvements. They run the experiments that establish and optimize deployment
|
|
patterns. They manage the drives and networking that allow us to serve hundreds
|
|
of petabytes at terabits per second, and their projections tell us that storage
|
|
must become bigger and faster.
|
|
|
|
Thinking like an operator, there are several areas where Swift should improve.
|
|
Replication and reconstruction continue to be a delicate balancing act between
|
|
moving data for the sake of expansion vs. ensuring durability. We need a better
|
|
handle on which data gets accessed, how frequently, and with what sort of
|
|
performance -- ideally indexed by tenant and even user. As more clusters are
|
|
stood up in more regions, we need to improve monitoring and recovery such that
|
|
clusters can run with minimal human intervention.
|
|
|
|
I look forward to working with the community to bring Swift into a future of
|
|
ever-larger and ever-more-numerous clusters.
|