{"thread":{"id":"22297","subject":"[OT] git fsck uncovers hardware/kernel problem?","startedAt":"2010-01-20T02:54:46Z","lastAt":"2010-01-20T02:54:46Z","messageCount":1,"participants":["John Koleszar"],"isPatch":false,"patchVersion":null,"patchTotal":null},"messages":[{"id":"132162","messageId":"4B567076.8030405@on2.com","threadId":"22297","inReplyTo":null,"subject":"[OT] git fsck uncovers hardware/kernel problem?","fromName":"John Koleszar","fromEmail":"john.koleszar@on2.com","sentAt":"2010-01-20T02:54:46Z","receivedAt":"2010-01-20T02:54:46Z","isPatch":false,"sender":{"key":"john.koleszar@on2.com","avatar":null},"body":"Hi all,\n\nI don't think that this is a git issue, but I've only reproduced it so \nfar using git and I know there are people here that can point me in the \nright direction, so please forgive the off-topic post.\n\nI was repairing a repo today that I thought was corrupt, and when I \nfinished, it fsck'd ok but then ran into another corruption error when \ngc'ing it. I realized these errors were intermittent. I ran into this \nsituation once before (same machine) and was time-pressed, so I rebooted \nand the problem went away. This time I have a little free time to \ninvestigate.\n\nIt looks like somehow my buffer cache is getting inconsistent with \nwhat's actually on the disk. I'm looking for input as to whether this is \na kernel problem, bad memory, dieing hard disk, etc. This machine has \nbeen stable, and I've had no unexplained crashes or other noticeable \ndata corruption. No errors from the kernel, no SMART errors. Only \nnotable thing about the repos I was working on when this happened is \nthat they were large -- exceeding the size of physical RAM. Lots of \nbinary objects. The one I'm looking at now is about 3.8GB packed.\n\nPlease see below -- any pointers would be appreciated, including to a \nmore appropriate list.\n\n$ git fsck --full\nerror: inflate: data stream error (incorrect data check)\nerror: corrupt loose object '060a8f0255a4a0b76197056e959bb0ce37b90dae'\nfatal: object 060a8f0255a4a0b76197056e959bb0ce37b90dae is corrupted\n\n$ echo 3 | sudo tee /proc/sys/vm/drop_caches\nPassword:\n3\n\n$ git fsck --fullerror: inflate: data stream error (incorrect data check)\nerror: corrupt loose object '0905cc800a08a24c8d83132c4d346d8b47403a73'\nfatal: object 0905cc800a08a24c8d83132c4d346d8b47403a73 is corrupted\n# Note this is a different object\n\n$ git fsck --fullerror: inflate: data stream error (incorrect data check)\nerror: corrupt loose object '0905cc800a08a24c8d83132c4d346d8b47403a73'\nfatal: object 0905cc800a08a24c8d83132c4d346d8b47403a73 is corrupted\n# but running again gives a consistent error until dropping caches again\n\n$ echo 3 | sudo tee /proc/sys/vm/drop_caches\n3\n\n$ git fsck --full\nerror: inflate: data stream error (incorrect data check)\nerror: corrupt loose object '3e9a88bfea886dd4bd4af69870e33f115bc80985'\nfatal: object 3e9a88bfea886dd4bd4af69870e33f115bc80985 is corrupted\n# fsck progresses much farther this time.\n\n$ sha1sum objects/3e/9a88bfea886dd4bd4af69870e33f115bc80985\nacffc0e434297e5115b2491101d0a0260a20a1aa  \nobjects/3e/9a88bfea886dd4bd4af69870e33f115bc80985\n\n$ echo 3 | sudo tee /proc/sys/vm/drop_caches\n3\n\n$ sha1sum objects/3e/9a88bfea886dd4bd4af69870e33f115bc80985\n83a1291ecb9e2fd9ded509aefba2a3c474d259c2  \nobjects/3e/9a88bfea886dd4bd4af69870e33f115bc80985\n# sha1sum is different, so this issue isn't limited to git\n\n$ git cat-file blob 3e9a88bfea886dd4bd4af69870e33f115bc80985>/dev/null\n# git's integrity checks pass on this file now.\n\n$ uname -a\nLinux cp-jk-linux 2.6.30-gentoo-r5 #3 SMP Wed Nov 4 12:53:38 EST 2009 \nx86_64 Intel(R) Core(TM)2 CPU 6600 @ 2.40GHz GenuineIntel GNU/Linux\n\n---\nThanks,\nJohn\n"}]}