> 1380610633.387730: POST / HTTP/1.1^M
> ThreadRPCServer method=getinfo
> CTxMemPool::accept() : accepted
> dc2941dd69b2f9fa2754f741dfba76489abef706bd237e3bd715181950723e4d (poolsz 1283)
> keypool reserve 15
> keypool return 15
> locktime : 5.996820 calltime : 0.000328 totaltime : 5.997148
>
I have found another circumstance where getinfo rpc function can take
more than 2 secs to return. Here are the circunstances:
received block 0000000000000013c889fbe1b2a23eb4b6bd38a8051420be1bf319da184f02ae
Committing 1015 changed transactions to coin database...
SetBestChain: new
best=0000000000000013c889fbe1b2a23eb4b6bd38a8051420be1bf319da184f02ae
height=261213 log2_work=72.38679 tx=24743665 date=2013-10-02 05:26:49
progress=0.999999
{"method":"getinfo","params":[],"id":783}
ThreadRPCServer method=getinfo
keypool reserve 15
keypool return 15
locktime : 0.000001 calltime : 2.453037 totaltime : 2.453038
This time locking the contention isn't from cs_main, cs_wallet mutex
lock. I think that what happen here is
1. CWallet::SetBestChain is called
2. This trigger ThreadFlushWalletDB that will lock wallet db and flush
it
3. getinfo will have to wait that ThreadFlushWalletDB ends to complete
GetOldestKeyPoolTime().
Just throwing a couple of ideas here. Comments are welcome:
1.
Is closing the db and reopen it the only way to flush the db to disk?
Can it be done asynchonously? I don't know BDB much but probably closing
the DB involves some blocking fsync() call and given that this follow
1015 records writing into the txdb, it takes more time than it should
otherwise to return.
2.
I'm relatively new to bitcoin. I understand that in a recent past
everything was using BDB. Now that only the wallet seems to keep using
it, maybe some options could be removed to make BDB more lightweight.
DB_THREAD:
>From what I see, BDB API access is already well synchronized by
bitcoind. There is no obvious reason to ask BDB API to be thread safe.
DB_INIT_TXN:
Correct me if I'm wrong, but I suspect this subsystem initialization to
be a vestige of when txdb was in BDB.
3.
not sure at all about this one but I'm throwing the idea anyway.
Personnally, I have no used for 'keypoololdest'. I might experiment
without it
How about having an optional bool param to 'getinfo' that could have the
name 'lowlatency' or 'nodb' to skip info gathering having the potential
to make the call longer?
------------------------------------------------------------------------------
October Webinars: Code for Performance
Free Intel webinars can help you accelerate application performance.
Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
the latest Intel processors and coprocessors. See abstracts and register >
http://pubads.g.doubleclick.net/gampad/clk?id=60134791&iu=/4140/ostg.clktrk
_______________________________________________
Bitcoin-development mailing list
Bitcoin-development@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bitcoin-development
