-
Notifications
You must be signed in to change notification settings - Fork 157
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Incubating Program: Follower Read With Applied Index #86
Labels
incubating-request/feature
This is an Incubating Program request for a feature
status/LGT2
Indicates that a PR has LGTM 2.
Comments
LGTM |
LGTM |
1 similar comment
LGTM |
AndreMouche
pushed a commit
to AndreMouche/community
that referenced
this issue
Nov 6, 2020
Signed-off-by: Rustin-Liu <[email protected]>
Closed by pingcap/tidb#21094. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Labels
incubating-request/feature
This is an Incubating Program request for a feature
status/LGT2
Indicates that a PR has LGTM 2.
Follower Read With Applied Index
Describe the feature or project you want to incubate:
Summary
This RFC proposes a improvement about getting snapshot on followers, which is
if the read request carries an applied index, the peer can get snapshot locally
wihtout any commucation with its leader, if only it has applied to the given
index. It's useful to reduce latency if the cluster is deployed in multi
datacenters.
Motivation
For clusters deployed in multi datacenters, the system latency could mainly
depend on the network RTT between datacenters. For example, suppose PDs are
deployed in Beijing, and TiKVs are deployed in Beijing and Xian (for high
availability). If a client which is near to Xian wants to read a region, it
needs to get a transaction timestamp from PDs (in Beijing), and then sends
requests to TiKVs in Beijing or Xian. For the latter case, TiKVs in Xian will
send read index requests to their leaders (in Beijing) internally, which still
involves a RTT crossing datacenters.
So, If we can add some proxies in the major datacenter, and let it help TiDBs
in Xian to get transaction timestamp and applied indices of all target regions,
the read latency between multi datacenter will be reduced from 2 RTT to 1 RTT.
Your RFC/Proposal?
tikv/rfcs#32
tikv/rfcs#33
Tracking list
TiKV tasks about follower read with applied index
TiDB tasks about follower read with applied index
Tasks about follower replication
The text was updated successfully, but these errors were encountered: