Tags: tikv/pd
Tags
gc_state_manager: Workaround mutex contention in high-concurrency sce… …nario (#10615) ref #10607 This PR provides a simpler workaround to the issue that when PD receives too many GC-related requests, it may suffer bad performace on the mutex of GCStateManager's mutex. In the follwing cases, the GCState will be fully loaded and inserted to a memory cache: * When GetGCState/GetAllKeyspacesGCState encounters a cache miss and reads the actual GCState from etcd * When AdvanceTxnSafePoint is called In the following cases, the GCState cache will be incrementally updated if it already exists in the cache: * When AdvanceGCSafePoint/SetGCBarrier/DeleteGCBarrier is called Also note that UpdateServiceGCSafePoint is internally redirected to one of AdvanceTxnSafePoint SetGCBarrier or DeleteGCBarrier. Signed-off-by: MyonKeminta <MyonKeminta@users.noreply.github.com>
server: add GC request metrics (#10610) server: add GC request metrics Signed-off-by: okjiang <819421878@qq.com>
chore: bump tidb-dashboard version (#10580) Signed-off-by: Suhaha <jklopsdfw@gmail.com>
*: support tso client suffixbit mechanism (#10340) * suffixbit Signed-off-by: tongjian <1045931706@qq.com> * revert some changes Signed-off-by: tongjian <1045931706@qq.com> * refactor Signed-off-by: tongjian <1045931706@qq.com> * lint Signed-off-by: tongjian <1045931706@qq.com> * revert Signed-off-by: tongjian <1045931706@qq.com> * remove local tso test Signed-off-by: tongjian <1045931706@qq.com> --------- Signed-off-by: tongjian <1045931706@qq.com>
PreviousNext