Overview of changes
- Enables high data density by turning multi-level indexing on by default. This allows efficient queries even where there is a lot of data per node.
- Core performance fixes to run better on lightweight (2 vCPU) machines
- Better handling of huge io spikes. The scenario is having many more requests from the clients than what the database cluster can handle.
- Tuning the Redis query layer performance
Key commits
- [
a4498b4] #88: Ensure we check TTL when we look up the type of a subdoc - [
beae94d] Remove Scoped watch stack from outbound calls. Gflags for consensus log - [
abe90b1] Make the appenders use a thread pool - [
6b33e8d] Bundle yb-prof.py in the release package. - [
e94574c] Change defaults for the number of concurrent tserver-to-master RPCs and Redis buffer size - [
e96545a] Allow std::atomic usage with clang and libstdc++ - [
13724f5] Remove unnecessary encoding and decoding from DocDB - [
b36c90e] Spawn a shell master when adding a new master. - [
702d528] Use multi-level indexes by default - [
72ffc21] Fixing reactor-test: DelayedTask::AbortTask should still call the callback for tasks that have not been scheduled - [
c0c4c7b] Add paging to table iterator to fix RaftConsensusITest.TestAutoCreateReplica