server:
- always has read & write timeouts
- ping handler never blocks the reader (see A)
- sends regular pings to check up on a client
A:
at some point server write buffer can become full, so in order not to
block reads from a client (see
https://github.com/gorilla/websocket/issues/97), server may skip some
pongs. As a result, client may disconnect. But you either have to do
that or block the reader. There is no third way.
client:
- optional read & write timeouts
- optional ping/pong to measure latency
save transactions to blockstore
move to a separate module
benchmark KVIndexer
batch write transactions
Benchmarks:
```
BenchmarkKVIndexerIndex-2 100000 516300 ns/op
PASS
ok github.com/tendermint/tendermint/blockchain/tx 56.506s
5,16 s for 10000 transactions
1 s for 2000 transactions
```
```
BenchmarkKVIndexerIndex-2 h 3000000 8622 ns/op
PASS
ok github.com/tendermint/tendermint/blockchain/tx 34.210s
86 ms for 10000 transactions
16 ms for 2000 transactions
```
```
BenchmarkKVIndexerIndex1-2 5000000 7160 ns/op
BenchmarkKVIndexerIndex500-2 20000 1750411 ns/op
BenchmarkKVIndexerIndex1000-2 10000 3573973 ns/op
BenchmarkKVIndexerIndex2000-2 5000 7836851 ns/op
BenchmarkKVIndexerIndex10000-2 1000 33438980 ns/op
PASS
ok github.com/tendermint/tendermint/blockchain/tx 209.482s
7,8 ms for 2000 transactions
```
[state] write test for ApplyBlock
review comments
- move txindexer to state
- fix type
save Tx Index as well
do not store tx itself in the result