Assertion error: start_block == end_block - 1

[tarlano]

I don't know how to reproduce this situation, but my entire infinite volume, on all devices in the network, stopped accepting writes today.

From the error log I found the following assertion error:

[        elle.assert         ] [fuse loop] assertion 'start_block == end_block - 1' failed at /builds/infinit/memo/src/infinit/filesystem/FileHandle.cc:385
[        elle.assert         ] [fuse loop]   #0  0x000000007fffffff: elle::AssertError::AssertError(char const*, char const*, unsigned long) +0x571
                                             #1  0x000000007fffffff: elle::_abort(std::string const&, char const*, int) +0xa7
                                             #2  0x000000007fffffff: infinit::filesystem::FileBuffer::_write_multi_multi(infinit::filesystem::FileHandle*, elle::ConstWeakBuffer, long, int, int) +0x125
                                             #3  0x000000007fffffff: infinit::filesystem::FileBuffer::write(infinit::filesystem::FileHandle*, elle::ConstWeakBuffer, unsigned long, long) +0x29f
                                             #4  0x000000007fffffff: infinit::filesystem::FileHandle::write(elle::ConstWeakBuffer, unsigned long, long) +0x2c
                                             #5  0x000000007fffffff: ???
                                             #6  0x000000007fffffff: fuse_fs_write_buf +0x230
                                             #7  0x000000007fffffff: ???
                                             #8  0x000000007fffffff: ???
                                             #9  0x000000007fffffff: ???
                                             #10 0x000000007fffffff: __fuse_process_cmd +0x17
                                             #11 0x000000007fffffff: elle::reactor::FuseContext::_loop_single() +0x1c3
                                             #12 0x000000007fffffff: elle::reactor::Thread::_action_wrapper(std::function<void ()> const&) +0x249
                                             #13 0x000000007fffffff: ???
                                             #14 0x000000007fffffff: make_fcontext +0x21
terminate called after throwing an instance of 'elle::AssertError'
  what():  assertion 'start_block == end_block - 1' failed at /builds/infinit/memo/src/infinit/filesystem/FileHandle.cc:385

The version is
$ infinit --version
0.8.0

Thanks,
Tony

[tarlano]

I had the volume block_size set to a small 1KB. After increasing it back to the default 1.04MB I am seeing better performance and possibly it will alleviate the block overrun in the memo FileHandler method.

To reproduce this issue you maybe able to set the block_size to small 1K (1024) and write several large files.

[Dimrok]

Hi @tarlano.

It's a known issue, you shouldn't use a block size smaller than 64KB (source @mnottale) for now (I'll add that to the documentation).
You can't write blobs of data larger than twice the block size (source @mnottale).

By experimenting, I found that, with my fuse configuration:

• I wasn't able to reproduce with block_size bigger than 2KB.
• The limit of data you can write with a block size of 1024 is 3024.

Here's a minimal test case to reproduce.

set -x

rm -rf .local

export MOUNTPOINT="$PWD/mountpoint"
export BLOCK_SIZE=2

infinit user create
infinit silo create filesystem silo
infinit network create --silo silo network
infinit volume create --network network --block-size $BLOCK_SIZE volume
infinit volume mount --name volume --mountpoint "$MOUNTPOINT" --cache --allow-root-creation &

sleep 1

DATA_SIZE=$(expr $BLOCK_SIZE \* 2)
echo $(printf '=%.0s' $(eval echo "{1..$DATA_SIZE}"))  >> "$MOUNTPOINT/data"