Skip to content
Commits on Oct 14, 2015
  1. @borshop

    Merge pull request #222 from basho/feature/bch/upgrade-cuttlefish-to-…

    …2.0.5
    
    Update cuttlefish to 2.0.5
    
    Reviewed-by: cuyler
    borshop committed Oct 13, 2015
Commits on Oct 13, 2015
  1. @javajolt

    Update cuttlefish to 2.0.5

    javajolt committed Oct 13, 2015
Commits on Sep 10, 2015
  1. @slfritchie

    Merge pull request #220 from basho/jhc-typo

    Correct format string/data mismatch
    slfritchie committed Sep 10, 2015
  2. @joecaswell

    Correct format string/data mismatch

    This was likely a typo from a long time ago.
    joecaswell committed Sep 9, 2015
Commits on Jun 18, 2015
  1. @borshop

    Merge pull request #216 from basho/bugfix/file-delete-blocking

    Prevent one frozen dir from blocking other deletes
    
    Reviewed-by: mrallen1
    borshop committed Jun 18, 2015
  2. @borshop

    Merge pull request #214 from basho/bugfix/avoid-fold-crash-on-invalid…

    …-key
    
    Avoid fold crash on invalid key
    
    Reviewed-by: mrallen1
    borshop committed Jun 18, 2015
  3. @engelsanchez

    Prevent one frozen dir from blocking other deletes

    Customers are seeing a problem where a single Bitcask instance that's
    folding blocks all others from completing their deferred deletes. This
    change makes the worker go through all pending deletes and finishing all
    the ready ones instead of stopping at the first one that is not ready.
    engelsanchez committed Jun 18, 2015
  4. @engelsanchez

    Add failing test for blocked file deletes

    This test shows that a bitcask directory that's frozen and therefore has
    its file deletions delayed after a merge blocks another directory from
    having its own files deleted, even if that other directory is finished
    folding and deletes are possible.
    engelsanchez committed Jun 18, 2015
  5. @borshop

    Merge pull request #213 from basho/bugfix/avoid-loading-keys-from-cor…

    …rupted-objects
    
    Avoid loading keys from corrupted entries
    
    Reviewed-by: mrallen1
    borshop committed Jun 18, 2015
  6. @engelsanchez

    Removing dead code after skipping removed

    Dialyzer complains, rightly, that this code is never touched after the
    logic changed so that we never skip binaries in order to validate every
    record's CRC.
    engelsanchez committed Jun 18, 2015
  7. @engelsanchez
Commits on Jun 17, 2015
  1. @engelsanchez

    Avoid crash on key transformation

    If the user defined key transformation function fails on a key read from
    disk, do not crash the process. Instead, keep going and log an error.
    engelsanchez committed May 19, 2015
  2. @engelsanchez

    Add failing test for key conversion errors

    When loading keys from a file, a user defined transformation function
    may be applied. If this function fails, currently we crash. But we want
    to be more robust than that. This test shows the crash on load and
    merge.
    engelsanchez committed May 19, 2015
Commits on May 19, 2015
  1. @engelsanchez

    Avoid loading keys from corrupted entries

    Due to an optimization added last year, when keys are loaded from a data
    file due to a bad or missing hintfile, keys were loaded even if the
    record did not pass the CRC test. This was done to avoid extra I/O by
    skipping reading the value, but it is incorrect and has caused a few
    customers some problems either by Bitcask crashing on startup or data
    scans failing later when hitting a corrupt key.
    engelsanchez committed May 19, 2015
  2. @engelsanchez

    Add failing test: key from corrupted entry loaded

    This test messes with a bitcask data value to make the CRC check fail.
    It demonstrates that due to an optimization, even if the CRC check fails
    and the entry is invalid and should be ignored, it still loads the key.
    engelsanchez committed May 19, 2015
Commits on Feb 25, 2015
  1. @borshop

    Merge pull request #210 from basho/the-great-versioning

    Pin cuttlefish to tag 2.0.1 explicitly
    
    Reviewed-by: javajolt
    borshop committed Feb 25, 2015
  2. @borshop

    Merge pull request #211 from basho/errors

    Remove incorrect option.
    
    Reviewed-by: mrallen1
    borshop committed Feb 25, 2015
  3. @cmeiklejohn

    Remove incorrect option.

    Both gcc and clang do not support a -errors flag, so remove it to
    prevent errors about using an undefined flag.
    cmeiklejohn committed Feb 25, 2015
  4. @mrallen1
  5. @borshop

    Merge pull request #208 from basho/17

    Add support for Erlang 17.
    
    Reviewed-by: andrewjstone
    borshop committed Feb 24, 2015
  6. @cmeiklejohn

    Add support for Erlang 17.

    cmeiklejohn committed Feb 24, 2015
Commits on Feb 21, 2015
  1. @borshop

    Merge pull request #207 from basho/meck

    Bump meck dependency on develop to match folsom.
    
    Reviewed-by: zeeshanlakhani
    borshop committed Feb 21, 2015
  2. @cmeiklejohn

    Bump meck dependency on develop to match folsom.

    The folsom dependency on develop is now at the 0.8.2 tag of meck;
    advance the bitcask tag to allow riak_kv and yokozuna to compile.
    cmeiklejohn committed Feb 21, 2015
Commits on Oct 17, 2014
  1. @borshop

    Merge pull request #201 from basho/merging-1.7.1

    Merging 1.7.1
    
    Reviewed-by: slfritchie
    borshop committed Oct 17, 2014
Commits on Sep 30, 2014
  1. @engelsanchez
Commits on Sep 17, 2014
  1. @borshop

    Merge pull request #198 from basho/bugfix/list-files-race

    Fix race listing readable files
    
    Reviewed-by: slfritchie
    borshop committed Sep 17, 2014
  2. @borshop

    Merge pull request #197 from basho/bugfix/extra-tombstones-on-update

    Fix extra tombstones on update
    
    Reviewed-by: slfritchie
    borshop committed Sep 16, 2014
Commits on Sep 16, 2014
  1. @engelsanchez

    Fix race listing readable files

    A race existed as demonstrated by the newly added
    list_data_files_race_test test. When listing readable files from a
    Bitcask directory for merging and such, the current write or merge
    output files should never be listed along with the other files. This
    could lead to merges trying to read active files while they are being
    written.
    
    Now, the code fetches the current merge and write files, lists the
    directory contents, filtering those out, and checks again if the merge
    and write file have changed. If they have, we try again, only succeeding
    if no changes have happened in between.
    engelsanchez committed Sep 16, 2014
  2. @engelsanchez

    Add failing list data files race test

    This test shows what happens when listing the readable files from a
    Bitcask directory. The list should not include the active write or merge
    files. But if those change after the code fetches the name of the
    current one, the listing will contain the new active write and/or merge
    files, skipping the *previous* write or merge file. That would be bad.
    
    The test uses meck to return the filename, then in one go write a new
    one to simulate the race.
    engelsanchez committed Sep 16, 2014
Commits on Sep 15, 2014
  1. @engelsanchez

    Fix extra update tombstones

    An unrelated fix in commit 4ab96d4 caused updates replacing a value
    in the same file to write extra tombstones that serve no purpose.
    Updates are supposed to create a tombstone for previous values only
    when the previous value was stored in an older file. Multiple
    values for the same key in the same file do not require these extra
    tombstones.
    engelsanchez committed Sep 15, 2014
  2. @engelsanchez

    Add failing update tombstones test

    Our current delete scheme requires that updates write tombstones for
    previous values when the previous value is in an file logically older
    than the current write file.  An unrelated change broke that logic, and
    now tombstones are written every single time a value is updated.
    
    This test demonstrates this by counting the number of tombstones
    expected for repeatedly updating the same key when two data files are
    created. Only a tombstone is required, but we get one per update.
    engelsanchez committed Sep 15, 2014
Commits on Sep 12, 2014
  1. @borshop

    Merge pull request #192 from basho/bugfix/expiry-remove-race

    Fix remove expired on read race
    
    Reviewed-by: slfritchie
    borshop committed Sep 12, 2014
  2. @borshop

    Merge pull request #190 from basho/bugfix/scan-error-deadlock

    Fix scan error deadlock
    
    Reviewed-by: slfritchie
    borshop committed Sep 12, 2014
  3. @borshop

    Merge pull request #189 from basho/bugfix/list-dir-port-refresh

    Refresh efile port if gone
    
    Reviewed-by: slfritchie
    borshop committed Sep 12, 2014
Commits on Sep 9, 2014
  1. @engelsanchez

    Fix remove expired on read race

    If an expired entry is found in they keydir on a read operation, the
    code immediately and unconditionally deletes it.  But the entry could
    have been updated since the read, and we could end up deleting a new
    value.
    
    This commit changes the delete to be conditional. It will succeed only
    if we are deleting exactly the same entry we found. If a newer entry is
    found instead, the get operation is retried.
    engelsanchez committed Sep 9, 2014
Something went wrong with that request. Please try again.