During some long run stability tests, I found there are low chances that Transaction::scan request will hang. After add some tracing logs (see https://github.com/pingyu/client-rust/tree/trace-scan), it show that the hang position is at the get_timestamp of pd client.
The tracing logs of a normal request (some irrelevant spans are ignored):
During some long run stability tests, I found there are low chances that
Transaction::scan
request will hang. After add some tracing logs (see https://github.com/pingyu/client-rust/tree/trace-scan), it show that the hang position is at theget_timestamp
of pd client.The tracing logs of a normal request (some irrelevant spans are ignored):
The abnormal request:
We can see that in the abnormal process, it did not enter the
TimestampOracle::get_timestamp
.