4 ms·

> Modern file systems use ordering points to maintain consistency in the face of system crashes. Could someone shed more light? In modern or older O/S, what e

> Modern file systems use ordering points to maintain consistency in the face of system crashes. Could someone shed more light? In modern or older O/S, what exactly happens during the crash?

tyingq 6y ago
I'd put it in 3 buckets of "not journaled", "metadata only journaled", and "data and metadata journaled". In the "not journaled" bucket, you have to hope something like fsck can make sense of the state of the filesystem. Often, the filesystem could be so "fscked" that you can't mount/boot. In the "metadata only journaled" bucket, if you use order...meaning writing the data before you write the metadata, you will have a consistent filesystem that can be mounted, albeit possibly missing some data. In the last bucket, you're also getting a consistent filesystem that can be mounted, but with less data loss. At the cost of some performance. I think they are saying both buckets 2 and 3 are "modern". I certainly encountered lots of fsck giving up in the 90's on "bucket 1" type filesystems.
- jzer0cool 6y ago
  Thanks! Is bucket 1 also related to hard drive won't boot. Eventually be able to recover mbr and proper boot location on and to the OS?
  - tyingq 6y ago
    Yes, "won't boot" meaning the root filesystem isn't recoverable, can't fully boot without that. You could have, for example, a working /boot, but a dead /. In that case, the kernel will boot up and then panic when it can't mount root. A bad mbr is also possible, for reasons unrelated to any filesystem. And a journaling filesystem doesn't always help with drive errors, etc. I was trying to scope down to filesystem issues due to an unorderly shutdown, where the disk drive itself is fine.
- j16sdiz 6y ago
  There are softupdate filesystem too. They are less common now, but still exist. https://en.wikipedia.org/wiki/Soft_updates https://en.wikipedia.org/wiki/Soft_updates
  - tyingq 6y ago
    Ah, yes, good callout. That reminded me of an LWN article on the topic: https://lwn.net/Articles/339337/ https://lwn.net/Articles/339337/

> Modern file systems use ordering points to maintain consistency in the face of system crashes. Could someone shed more light? In modern or older O/S, what exactly happens during the crash?

tyingq 6y ago

I'd put it in 3 buckets of "not journaled", "metadata only journaled", and "data and metadata journaled". In the "not journaled" bucket, you have to hope something like fsck can make sense of the state of the filesystem. Often, the filesystem could be so "fscked" that you can't mount/boot. In the "metadata only journaled" bucket, if you use order...meaning writing the data before you write the metadata, you will have a consistent filesystem that can be mounted, albeit possibly missing some data. In the last bucket, you're also getting a consistent filesystem that can be mounted, but with less data loss. At the cost of some performance. I think they are saying both buckets 2 and 3 are "modern". I certainly encountered lots of fsck giving up in the 90's on "bucket 1" type filesystems.

jzer0cool 6y ago
Thanks! Is bucket 1 also related to hard drive won't boot. Eventually be able to recover mbr and proper boot location on and to the OS?
- tyingq 6y ago
  Yes, "won't boot" meaning the root filesystem isn't recoverable, can't fully boot without that. You could have, for example, a working /boot, but a dead /. In that case, the kernel will boot up and then panic when it can't mount root. A bad mbr is also possible, for reasons unrelated to any filesystem. And a journaling filesystem doesn't always help with drive errors, etc. I was trying to scope down to filesystem issues due to an unorderly shutdown, where the disk drive itself is fine.
j16sdiz 6y ago
There are softupdate filesystem too. They are less common now, but still exist. https://en.wikipedia.org/wiki/Soft_updates https://en.wikipedia.org/wiki/Soft_updates
- tyingq 6y ago
  Ah, yes, good callout. That reminded me of an LWN article on the topic: https://lwn.net/Articles/339337/ https://lwn.net/Articles/339337/