(DNM) D4N dirty prefix removal #20
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
into bucket directory, object directory and then versions representing file names. This commit for extracts bucket id, object name, version, offset, and length from input key and organizes data into directories - with bucket id, object name being used to create directories and filenames within them using version and offset and length. Data can be restored back to in memory LFUDA and dirty object data structure once rgw is restarted. Updating unit tests. Signed-off-by: Pritha Srivastava <[email protected]>
samarahu
force-pushed
the
d4n-dirty-prefix-removal
branch
from
ac85da1 to
062c198
Compare
samarahu
commented
Nov 4, 2024
samarahu
commented
Nov 4, 2024
samarahu
commented
Nov 4, 2024
samarahu
force-pushed
the
d4n-dirty-prefix-removal
branch
from
062c198 to
cc2beac
Compare
for unversioned buckets. 1. d4n/directory: Update logging 2. d4n/filter: Update `set_head_obj_dir_entry` to handle null versioning and update `prevVersion` for versioned head objects 3. rgw/d4n: Simplify code to only support unversioned bucket deletes Signed-off-by: Samarah <[email protected]>
clean (non-dirty) objects for delete_obj method. 1. rgw/d4n: miscellaneous fixes that include modifying set_head_obj_dir_entry() and delete_obj method to handle clean objects only. 2. rgw/d4n: changes to design when write cache is turned off: a. do not store a hash entry for latest version for versioned buckets, the latest version will always be fetched from backend store, to maintain correct order of versions during a delete object request. Signed-off-by: Pritha Srivastava <[email protected]>
and BlockDirectory. Adds methods to add, list a range and remove members from a redis ordered set ordered by score passed in. Also adds unit test cases to test the methods. Signed-off-by: Pritha Srivastava <[email protected]>
WATCH, EXEC, INCR, MULTI and DISCARD redis commands. 1. d4n/directory: support for watch, exec and incr methods in Block and Object Directory. 2. d4n/directory: support for redis MULTI and DISCARD command, adding a test case to exercise usage of MULTI. Signed-off-by: Pritha Srivastava <[email protected]>
operations atomic while modifying the latest and null hash entry in case of non-versioned buckets. Signed-off-by: Pritha Srivastava <[email protected]>
1. ordered set to maintain versions of a dirty object 2. creation of delete marker in case of a simple delete request 3. deletion of a specific version from the ordered set 4. cleaning method deletes from ordered set for dirty objects 5. use of redis atomicity constructs wherever needed Signed-off-by: Pritha Srivastava <[email protected]>
to lazy deletion - delete dirty objects data blocks in cleaning method and non-dirty objects data blocks in eviction method 1. rgw/d4n: Implement lazy deletion 2. rgw/d4n: cleaning method now supports delete markers also (both versioned and null). The delete markers are written correctly to the backend store. Also modifying the description of rgw_d4n_cache_cleaning_interval to explicitly state that the duration is in seconds. 3. rgw/d4n: do not call invalidate_dirty_object in case of a simple delete request for dirty objects belonging to a versioned bucket. In this case, a delete marker needs to be created instead of invalidating/deleting an object. 4. rgw/d4n: Update lazy deletion in policy Signed-off-by: Samarah <[email protected]> Signed-off-by: Pritha Srivastava <[email protected]>
samarahu
force-pushed
the
d4n-dirty-prefix-removal
branch
from
cc2beac to
3e82b54
Compare
filter driver and ssd backed cache. 1. rgw/d4n: addressing concurrency issues by adding a refcount to each block. Blocks with positive refcounts are pinned (similar to dirty blocks) and not eligible for eviction or deletion. Updating unit tests also. 2. rgw/cache: addressing concurrency issues while directories creation, deletion, updating xattrs, get_attr and put. Signed-off-by: Pritha Srivastava <[email protected]>
pritha-srivastava
force-pushed
the
wip-rgw-d4n-next
branch
from
66c58ca to
31ecd67
Compare
samarahu
force-pushed
the
d4n-dirty-prefix-removal
branch
from
3e82b54 to
67f2b59
Compare
samarahu
changed the title
| @@ -521,6 +514,13 @@ void D4NFilterObject::set_attrs_from_obj_state(const DoutPrefixProvider* dpp, op | |||
|
|
|||
| bl_val.append(this->get_bucket()->get_name()); | |||
| attrs[RGW_CACHE_ATTR_BUCKET_NAME] = std::move(bl_val); | |||
|
|
|||
| if (dirty) { | |||
There was a problem hiding this comment.
I have one generic comment for this PR: since you do not add an xattr for a clean block in handle_data(), lets do the same everywhere. We can avoid a call to cache by not setting an xattr in case of a clean block. So modify set_attrs_from_obj_state() to set the xattr for a head block only when it is dirty. The absence of the xattr signifies that the block is clean. Make bool dirty=false as the default argument to this method, so that you just need to pass in true only for dirty blocks. Also make changes while restoring data, only if the xattr is present and set to 1, then set that block to dirty, else set the block to clean.
There was a problem hiding this comment.
Also what is the status of testing of this PR? Have you tested these changes? and the data restoration also? Let me know.
There was a problem hiding this comment.
I have tested the PR with lazy deletion for both unversioned and versioned buckets (suspended and enabled), and I tested data restoration as well as restoration during a delete before the cleaning cycle. The data is handled correctly and the cache entries are deleted as expected during the cleaning cycles.
There was a problem hiding this comment.
Also please test 'get' requests, after a 'put' for both versioned and unversioned buckets if you haven't done so.
There was a problem hiding this comment.
You can also ensure that the 'dirty' xattr is set by using getfattr -d -- <filename>.
samarahu
force-pushed
the
d4n-dirty-prefix-removal
branch
3 times, most recently
from
4d0ac4c to
c0d5e9d
Compare
added 3 commits
November 14, 2024 15:37
Signed-off-by: Samarah <[email protected]>
Signed-off-by: Samarah <[email protected]>
write Signed-off-by: Samarah <[email protected]>
samarahu
force-pushed
the
d4n-dirty-prefix-removal
branch
from
c0d5e9d to
9e0bd2f
Compare
pritha-srivastava
force-pushed
the
wip-rgw-d4n-next
branch
from
31ecd67 to
a6980d5
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Includes commits to do the following:
delmethod's implementation from the RedisDriver and replacedelwith adeletecall in the RedisDriver unit test