Remove ThinClient from dos/#117
Conversation
gregcusack
force-pushed
the
remove-thin-client-from-dos
branch
from
21e39ec to
70324f5
Compare
gregcusack
added
CI
gregcusack
force-pushed
the
remove-thin-client-from-dos
branch
2 times, most recently
from
3edaa52 to
f2e0867
Compare
| let (rpc_addrs, tpu_addrs): (Vec<_>, Vec<_>) = nodes | ||
| .iter() | ||
| .filter_map(|node| node.valid_client_facing_addr(protocol, socket_addr_space)) | ||
| .unzip(); | ||
| let num_nodes = tpu_addrs.len(); | ||
| ( | ||
| ThinClient::new_from_addrs(rpc_addrs, tpu_addrs, connection_cache), | ||
| num_nodes, | ||
| ) |
There was a problem hiding this comment.
this seems to be spinning up rpc multiple clients under the wrapper of a ThinClient. I was able to get similar performance with a single TPU Quic client. This similar performance is why I changed it to a single TPU Client. However, I am not sure I have full context as to whether solana-dos needs to spin up multiple TPU clients for reasons other than maximizing rate_per_second
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #117 +/- ##
========================================
Coverage 81.8% 81.8%
========================================
Files 838 838
Lines 226389 226273 -116
========================================
- Hits 185342 185263 -79
+ Misses 41047 41010 -37 |
KirillLykov
left a comment
KirillLykov
left a comment
There was a problem hiding this comment.
looks good, the only thing is that this PR does two things:
- ThinClient -> TpuClient in dos
- removes ThinClient from bench-tps
I would split these changes because they are unrelated from what I see.
| let rpc_pubsub_url = format!("ws://{}/", cluster.entry_point_info.rpc_pubsub().unwrap()); | ||
| let rpc_url = format!("http://{}", cluster.entry_point_info.rpc().unwrap()); | ||
|
|
||
| let ConnectionCache::Quic(cache) = &*cluster.connection_cache else { |
There was a problem hiding this comment.
Ya tbh i thought so too lol. But it is used in ThinClient here, here, and here.
Also looks like it is a valid way to get a reference to the internal structure in an Arc. See: https://doc.rust-lang.org/beta/std/sync/struct.Condvar.html and https://stackoverflow.com/questions/62651479/understanding-to-access-a-rust-arc
| .unwrap(), | ||
| cluster.connection_cache.clone(), | ||
| )); | ||
| let rpc_pubsub_url = format!("ws://{}/", cluster.entry_point_info.rpc_pubsub().unwrap()); |
There was a problem hiding this comment.
myabe move this common code to create TpuClient to some helper function in the scope of tests submodule?
gregcusack
force-pushed
the
remove-thin-client-from-dos
branch
2 times, most recently
from
78622fb to
cf58be6
Compare
gregcusack
dismissed stale reviews from CriesofCarrots and KirillLykov
via
7d09b50
gregcusack
force-pushed
the
remove-thin-client-from-dos
branch
from
cf58be6 to
7d09b50
Compare
* remove `ThinClient` from `dos/` and replace `ThinClient` with `TpuClient` * remove test for valid_client_facing_addr since it is no longer used
…za-xyz#117) Minor updates after another sync run.
5 participants
5th PR on the way to remove ThinClient completely.
See
bench-tpsclient totpu-clientsolana-labs/solana#35335ThinClientand removeThinClientfrombench-tpssolana-labs/solana#35365get_clientandget_multi_client#177get_clientandget_multi_client#184Problem
ThinClientis deprecated, so we need to removesolana-dos's reliance on itSummary of Changes
Use
TpuClientinsolana-dosinstead ofThinClient