sc handler optimisations #1396
sc handler optimisations #1396
Conversation
| @@ -387,6 +422,7 @@ init(Args) -> | |||
| [ libp2p_swarm:listen(SwarmTID, Addr) || Addr <- ListenAddrs ]), | |||
| NewState = #state{swarm_tid = SwarmTID, blockchain = Blockchain, | |||
| gossip_ref = Ref}, | |||
| true = ets:insert(?CACHE, {?CHAIN, Blockchain}), | |||
There was a problem hiding this comment.
should be pretty easy to move this into a handle_continue/2 so we load the chain into the cache table before receiving any other messages post-init. would it make sense (perhaps later) to consolidate these "blockchain worker cache items" into a single shared globally readable cache, owned and updated by the worker, and consolidate the "current height" cache that's in review to merge into the txn mgr? any others doing similar work?
There was a problem hiding this comment.
we def need to get loading the chain out of the init, using handle_continue would work well there. I dont want to address that in this PR as its dramatically increases the impact and thus the risk.
We can address in a follow up PR and that could also include replacing all blocking calls to the worker to return the chain with the new non blocking call. Extending the cache to also include the height and additional data does make sense too.
andymck
force-pushed
the
andymck/optimise-core-sc-server-handler
branch
from
5036b49 to
831d4cc
Compare
andymck
force-pushed
the
andymck/optimise-core-sc-server-handler
branch
from
831d4cc to
ee9ae87
Compare
| @@ -506,6 +545,7 @@ handle_call({install_aux_snapshot, Snapshot}, _From, | |||
| {ok, GossipRef} = add_handlers(SwarmTID, NewChain), | |||
| notify({new_chain, NewChain}), | |||
| blockchain_lock:release(), | |||
| true = ets:insert(?CACHE, {?CHAIN, NewChain}), | |||
There was a problem hiding this comment.
don't think that we need this one here, this is for the aux ledger.
| @@ -538,16 +578,19 @@ handle_call({add_commit_hook, CF, HookIncFun, HookEndFun} , _From, #state{blockc | |||
| Ledger = blockchain:ledger(Chain), | |||
| {Ref, Ledger1} = blockchain_ledger_v1:add_commit_hook(CF, HookIncFun, HookEndFun, Ledger), | |||
| Chain1 = blockchain:ledger(Ledger1, Chain), | |||
| true = ets:insert(?CACHE, {?CHAIN, Chain1}), | |||
| {reply, Ref, State#state{blockchain = Chain1}}; | ||
| handle_call({add_commit_hook, CF, HookIncFun, HookEndFun, Pred} , _From, #state{blockchain = Chain} = State) -> | ||
| Ledger = blockchain:ledger(Chain), | ||
| {Ref, Ledger1} = blockchain_ledger_v1:add_commit_hook(CF, HookIncFun, HookEndFun, Pred, Ledger), | ||
| Chain1 = blockchain:ledger(Ledger1, Chain), | ||
| true = ets:insert(?CACHE, {?CHAIN, Chain1}), |
| {reply, Ref, State#state{blockchain = Chain1}}; | ||
| handle_call({remove_commit_hook, RefOrCF} , _From, #state{blockchain = Chain} = State) -> | ||
| Ledger = blockchain:ledger(Chain), | ||
| Ledger1 = blockchain_ledger_v1:remove_commit_hook(RefOrCF, Ledger), | ||
| Chain1 = blockchain:ledger(Ledger1, Chain), | ||
| true = ets:insert(?CACHE, {?CHAIN, Chain1}), |
| @@ -743,6 +787,7 @@ handle_info({'DOWN', RocksGCRef, process, RocksGCPid, Reason}, | |||
| {noreply, State#state{rocksdb_gc_mref = undefined}}; | |||
|
|
|||
| handle_info({blockchain_event, {new_chain, NC}}, State) -> | |||
| true = ets:insert(?CACHE, {?CHAIN, NC}), | |||
There was a problem hiding this comment.
pretty sure this is duplicative
There was a problem hiding this comment.
I'm worried about how many times we insert the chain into the cache. I've added some that I think are duplicated, but I think that someone should add logging to each of these inserts and make sure that we're only calling it once on any given code path. the important ones are:
- Startup with existing chain
- Load new genesis
- load snapshot
|
So the approach I took was to insert to the cache anytime the chain in state is being updated |
Provides optimisations/improvements of the state channel GRPC handler and the shared
blockchain_state_channel_commonmodule. This mod is shared across both grpc and libp2p transport layers for state channel requests.Optimisation/improvements include:
blockchainandledgerin the state channel handler state