Feature enhancement: drop some client requests if its in-queue time exceed client timeout

[hycdong]

Feature Request

When the client requests sent into Pegasus, it will firstly be in queue, then execution. When server load added, sometimes in-queue time may shortly increase. If the request's in-queue time exceed the client timeout(recorded in request header), those requests can be dropped, it will be timeout finnally, no need to be executed.

We plan to add this config for following RPC code:

RPC_RRDB_RRDB_PUT
RPC_RRDB_RRDB_MULTI_PUT
RPC_RRDB_RRDB_REMOVE
RPC_RRDB_RRDB_MULTI_REMOVE
RPC_RRDB_RRDB_DUPLICATE
RPC_RRDB_RRDB_GET
RPC_RRDB_RRDB_MULTI_GET
RPC_RRDB_RRDB_TTL

Those RPC codes are not add it:

// not_idempotent write task codes won't added
RPC_RRDB_RRDB_INCR
RPC_RRDB_RRDB_CHECK_AND_SET
RPC_RRDB_RRDB_CHECK_AND_MUTATE
// scan related rpc codes won't added
RPC_RRDB_RRDB_SORTKEY_COUNT
RPC_RRDB_RRDB_GET_SCANNER
RPC_RRDB_RRDB_SCAN
RPC_RRDB_RRDB_CLEAR_SCANNER

The following perf-counters will be added:

zion*profiler*RPC_RRDB_RRDB_PUT.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_MULTI_PUT.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_REMOVE.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_MULTI_REMOVE.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_DUPLICATE.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_GET.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_MULTI_GET.rpc.dropped
zion*profiler*RPC_RRDB_RRDB_TTL.rpc.dropped