log☇︎
36 entries in 0.469s
mircea_popescu: meanwhile in news : http://btcbase.org/log/2019-01-19#1888338 << mystery solved. trb uses bdb for block indexing ; the index had an incorrect lobe affecting some blocks ; whenever someone asked for one of those blocks, bdb died and trb never returned. ☝︎
mircea_popescu: http://btcbase.org/log-search?q=%22bdb%22+%22locks%22
a111: Logged on 2017-08-06 03:43 asciilifeform: my suspicion is that the bdb locks patch somehow has no effect when bdb built on netbsd
asciilifeform: but the results ( at least in asciilifeform's torture room ) have been disappointing, currently afaik nobody even knows why bdb ignores locks knob ( perma-hosing trb ) !! under bsd
asciilifeform: my suspicion is that the bdb locks patch somehow has no effect when bdb built on netbsd ☟︎
mircea_popescu: TomServo yeah it's required if you want to sync, bdb as shipped with original btc croaks on larger blocks
mircea_popescu: kakobrekla iirc it was the bdb locks thing fix ?
mircea_popescu: TomServo prolly the bdb locks issue.
assbot: [BTC-dev] (CORRECTED) Bullet in the Forehead for the BDB LocksIdiocy ... ( http://bit.ly/1JWNo0i )
BingoBoingo: http://log.bitcoin-assets.com/?date=03-01-2016#1359638 << Reason for this is known. I bizzaro BingoUniverse malloc makes space for the BDB based on max locks in advance. ☝︎
thestringpuller: mod6: was this started from a /clean/ env or is this trying to load up a blockchain from christmas past? << I completely nuked the datadir .bitcoin with rm -rf *; whenever I do an rpc call after the thing _tries_ to spin up, it borks. crashes without anything except what was in the db.log. I'm almost certain it's an issue with bdb because no blocks are downloaded or anything. cause it connects to the node I specify gets basic
BingoBoingo: ben_vulpes: LibreSSL node also runs with ~1GB of ram utilized at startup after last BDB locks patch because OpenBSD malloc something or other
ascii_field: as in, didn't have the bdb locks fix patch applied ?
assbot: Logged on 20-09-2015 09:36:21; BingoBoingo: On OpenBSD (non-pogoable) the latest BDB locks fix added ~800MB of memory allocation for bitcoind
BingoBoingo: On OpenBSD (non-pogoable) the latest BDB locks fix added ~800MB of memory allocation for bitcoind ☟︎
mircea_popescu: asciilifeform i can actually unwedge it. i am satisfied the problem here is a subtle memory issue (not directly related to the bdb locks thing). specifically, to verify block 367851 bitcoind needs a certain amount of CONTIGUOUS memory. but it doesn't know this.
BingoBoingo: joecool: With the anticipated spam attacks ideally you will want to bolt the Ophanage amputation and thermonuke modifications to whatever you choose. Also make sure you fix the BDB locks problem in your chose codebase. Anything post 0.3.21 and pre 0.8 should be fine. If you don't have particular things you fuss over 0.5.4'stator/rotor' is probably as close to idea as you will come
assbot: [BTC-dev] (CORRECTED) Bullet in the Forehead for the BDB LocksIdiocy ... ( http://bit.ly/1JWNo0i )
mircea_popescu: http://log.bitcoin-assets.com/?date=01-08-2015#1221203 << i don't think you understand what "locks" means in bdb parlance ☝︎
assbot: [BTC-dev] (CORRECTED) Bullet in the Forehead for the BDB LocksIdiocy ... ( http://bit.ly/1JWNo0i )
assbot: [BTC-dev] (CORRECTED) Bullet in the Forehead for the BDB LocksIdiocy ... ( http://bit.ly/1JWNo0i )
assbot: [BTC-dev] Bullet in the Forehead for the BDB Locks Idiocy ... ( http://bit.ly/1MYQuE4 )
BingoBoingo: <asciilifeform> ph0rk ?! << BDB exhaustion, upping maxlocks and DBobjects fixes
punkman: you gotta read BDB code for those locks
ascii_field: and was forgotten when bdb locks patch discovered
mircea_popescu: in general, going by the "never one single bug in kitchen, and noting that bdb locks are a resource of the kind that pointers are, just one easier to exhaust, i propose this as likely, on the basis of my business heuristics.
mod6: BingoBoingo: cool. ive got openssl/bdb/boost built now on obsd 5.6, then just gotta apply sublte obsd changes to v0.5.3.1 and then try the static build. if all works & pulls blocks, I'll make patches: one for the v0.5.3.1 source and one for `auto.sh' (which needs a few tweaks). maybe you can just point at that instead. anyway, no worries.
asciilifeform: not even the bdb locks patch
assbot: Check effective maximum bdb locks (possibly overridden in DB_CONFIG),... - Gitorious ... ( http://bit.ly/1577zv1 )
mod6: at first, I just set max locks to 40000, and nothing else, but ended up with liek 265Gb of BDB Tx logs
mod6: I was able to get upto the current block 2 different times with patches 1->5 (which includes ascii's patches, ben's patch, and my removal of checkpoints) with two seperate bdb configs. The first one just changed this: dbenv.set_lk_max_locks(40000); But ended up with a huge amount of BDB tx log files.
mircea_popescu: this is useful if you keep stopping every x blocks, backuping bdb files
mircea_popescu: asciilifeform bdb needs a set_lk_max_locks directive to function properly with larger blocks.
asciilifeform: mircea_popescu: bdb locks initialisation issue << my current study suggests malformed txes in the mempool also.
mircea_popescu: asciilifeform: does anyone know the cause of permanent blockchain wedging in 0.5.3 ? << i am pretty certain it's the bdb locks initialisation issue.
jborkl_: people were all up in arms about having 80 to 1000 unconfirmed transactions waiting. with NO testing, people made larger blocks - that broke BDB