[BOUNTY] Help test version 0.8, blockchain upgrade - Windows XP with an NTFS filesystem #2164

I tried editing test file in forked QA but nothing happens except progress meter spinning around the cat forever, so I'll drop everything related to test here. It is way too late to mess with github now!

Test the upgrade-to-release-0.8 code - Windows XP with an NTFS filesystem, upgrade from v0.7.1-beta

CPU AMD Athlon 64X2 Dual Core 4800+, Chipset AMD 780G/SB700, 2 GB RAM, HDD 74 GB WD Raptor 10k RPM SATA, Swap file 4 GB fixed size, Windows XP SP3 latest build.

Bitcoin-qt runs as server, using empty wallet.

Bitcoin datadir before test started, client synced with network and shutdown, debug.log deleted = 5,92 GB Bitcoin datadir after reindexing blocks = 11,1 GB Directories created and their sizes after reindexing blocks:

/blktree 29.7 MB /blocks 5.06 GB /coins 143 MB

Bitcoin-qt.exe process CPU usage while reindexing blocks = mostly 50%, peaks at 90% Bitcoin-qt.exe process RAM usage while reindexing blocks = 40 MB to 120 MB, peaks at 200 MB HDD activity while reindexing blocks = minimal

Noticable reindexing slowdown at around 25k and 60k blocks to go. Major reindexing slowdown at around 5.5k blocks to go. It seems mentioned slowdowns happen while other nodes on network pull blockchain data. My connection max upload speed is just 32 kB. "Bytes sent" for bitcoin-qt on my firewall tend to go crazy and abruptly jump to impossible values, like 100 MB uploaded after 10 or so minutes, which is not even remotely possible. Neither Terracoin nor Freicoin clients have same or similar issue.

At around 3k block to go I shutdown bitcoin-qt.exe in attempt to get rid off leechers, clueless = bitcoin-qt exited with error:

http://i.imgur.com/ybYWO.png

Bitcoin-qt restarted, reindexing not speeding up even a bit so I shutdown it again = same error as previously. Computer restarted, including ADSL modem, bitcoin-qt started again = same slow reindexing, crash after few minutes with error:

http://i.imgur.com/whzjF.png

Another start, slow reindexing but no crash = reindexing completed, bitcoin-qt shutdown at block 215.911 One more start, just to check for errors = all fine, bitcoin-qt shutdown at block 215.912

Log file:

http://www.sendspace.com/file/twtwsa

I've never seen the first error message (that on screenshot 1), it that really a OS dialog? Looks a little weird. Is your PC uo to date in terms of Windows updates? You have XP SP3 and all recent security and Visual C++ Runtime updates installed?

Dialog is produced by dwwin.exe (DrWatson) crash reporter, file scanned just to make sure it is clean:

https://www.virustotal.com/file/92e5974dfd91acebf5d8bd5f14361c0afd7528ef6503d1d8a8c26e64c115a0cb/analysis/1357828251/

All OS patches found here already applied:

http://www.microsoft.com/en-us/download/windows-xp.aspx?q=windows%20xp&fs=applicableproducts~%255eWindows%3bapplicableproducts~%255e%2522Windows%252fWindows%2520XP%2522%3bproducttype~%2522Security%2520Patch%2522&p=1&r=10&t=463&s=availabledate~Descending

Test run number 2:

The same behaviour as 1st run, this time without Visual C++ runtime error. Bitcoin-qt started and shutdown by me like 10 times after running from few up to 15 minutes = OSS GUI Client error seems to appear on random. Over time, debug.log file becomes ruined. Linked bellow is last one, created on upgrade start and not touched until test finished:

http://www.sendspace.com/file/4rzxwp

Reindexing slowdowns at exactly the same blockchain heights as previously. Test aborted at around 5k blocks to go. I will run one more test, this time not touching bitcoin-qt at all, to see if it can finish reindexing without user interference and hopefully get uncorrupted debug.log file.

How much RAM has your machine?

You didn't finish the test plan-- I only see the new version of bitcoin running in your debug.log files, the last step in the test plan is to make sure downgrading works properly.

The idea behind forking the test plan and then actually editing it as you test is to treat it as a checklist, so testers don't accidentally forget to test something that needs testing.

Also: did all of the crashes happen at shutdown?

Test run number 3:

Upgrade Note how much disk space is being used by the bitcoin data directory. = 5,9 GB Run the new bitcoind/Bitcoin-Qt binary. = Done EXPECT (all tests except for FAT32 filesystem): blktree/ subdirectory created in bitcoin data directory = Yes small (less than 1GB) increase in disk space usage = No, datadir size almost doubled (check first post) startup takes longer than ususual (blocks are re-indexed) = Yes but startup does not take hours (blocks are not re-downloaded) = Completed after 4 hours 25 min

Log file:

http://www.sendspace.com/file/x6aaw6

Downgrade Shut down the new bitcoind/Bitcoin-Qt, wait for it to completely exit = Done Run the old version of bitcoind/Bitcoin-Qt = Done EXPECT: Quick startup = Same speed as before this test Old version is synchronized to blockchain from where it last exited = Yes

Log file:

http://www.sendspace.com/file/le4g9i

Older version of bitcoin-qt ruined debug.log file created by newer version, hence two files uploaded, one for each phase. In this test run, I let newer bitcoin-qt version complete reindexing without interfering with it in any way. Once reindexing of older blocks and syncing with newer ones was completed, shutdown of newer bitcoin-qt resulted in no error message.

@ Diapolo - Machine has 2 GB RAM, more specs in my first post here @ Gavin - OSS GUI Client crashes happen randomly on user-initiated shutdown before reindexing is completed. Visual C++ crash happened just once in all 3 runs, while bitcoin-qt was reindexing.

Perhaps we have a problem when aborting -reindex then? I'll try a -reindex later... @gavinandresen I also got a MSVC++ Runtime error message, while the -reindex was running. That was with an own build... restarting the client without -reindex just leads to the same crash. Will need to start an IBD I guess.

Regarding datadir size, new version of client creates /blocks subdirectory, copies existing blockchain and index files there - or it just creates dummy files of same size, can't tell - and never deletes them, hence size of datadir almost doubling after full reindex.