diff options
| author |  Eric Anderson <ejona@google.com> | 2016-04-29 13:55:18 -0700 |
|---|---|---|
| committer | Eric Anderson <ejona@google.com> | 2016-04-29 13:55:18 -0700 |
| commit | 90b4a87d3ba78326d7ff523b3412684aa905670e (patch) | |
| tree | e600796b03d116dabe01bca217be617ad7f86785 | |
| parent | f921f2bee1e5a24f0259868ce0a622abfa24d331 (diff) | |
| parent | 2a932daaafbec9874d9fb594419f68f9cb8f9187 (diff) | |
Merge pull request #6372 from ctiller/fail_fast
Spell out fail fast semantics
| -rw-r--r-- | doc/fail_fast.md | 15 |
1 files changed, 15 insertions, 0 deletions
diff --git a/doc/fail_fast.md b/doc/fail_fast.md new file mode 100644 index 0000000000..3ed4297194 --- /dev/null +++ b/doc/fail_fast.md @@ -0,0 +1,15 @@ +gRPC Fail Fast Semantics +======================== + +Fail fast requests allow terminating requests (with status UNAVAILABLE) prior +to the deadline of the request being met. + +gRPC implementations of fail fast can terminate requests whenever a channel is +in the TRANSIENT_FAILURE or SHUTDOWN states. If the channel is in any other +state (CONNECTING, READY, or IDLE) the request should not be terminated. + +Fail fast SHOULD be the default for gRPC implementations, with an option to +switch to non fail fast. + +The opposite of fail fast is 'ignore connectivity'. + |