librbd: Asynchronous (v2) image creation #9585
Conversation
|
@dillaman I'm investigating the failure, but the patch should mostly be reviewable. |
| @@ -0,0 +1,633 @@ | |||
| // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- | |||
There was a problem hiding this comment.
Minor: move to "operation" subdirectory / namespace
| @@ -149,6 +149,18 @@ namespace librbd { | |||
| parent); | |||
| } | |||
|
|
|||
| void create_image(librados::ObjectWriteOperation *op, uint64_t size, uint8_t order, | |||
There was a problem hiding this comment.
I'd leave all the sync versions that are still needed by librbd -- just have the sync version call the async version to setup the object operation.
There was a problem hiding this comment.
The sync versions do use the async helpers which got introduced in the prior commit. This commit changes the sync version to use async helpers.
|
|
||
| delete m_journaler; | ||
|
|
||
| if ((m_r_saved == 0) || (*result < 0)) { |
There was a problem hiding this comment.
@dillaman Here, the cleanup cuts short if an error in encountered when shutting down journaler. Would it makes sense to continue deleting journal objects in this case?
There was a problem hiding this comment.
@vshankar Yeah, I can't think of a reason why not to continue on.
|
@ceph-jenkins retest this please |
|
Change in this patchset:
|
| using klass = RemoveRequest<I>; | ||
| Context *ctx = create_context_callback<klass, &klass::handle_init_journaler>(this); | ||
|
|
||
| m_journaler->init(ctx); |
There was a problem hiding this comment.
Check for exist before initializing the journal.
There was a problem hiding this comment.
I kind of relied on JournalMetadata::init which already calls aio_watch* on the (header) object which should return -ENOENT. Shouldn't that suffice?
[*] https://github.com/ceph/ceph/blob/master/src/journal/JournalMetadata.cc#L449
There was a problem hiding this comment.
-ENOENT can also be returned if the client isn't registered. If you stat it first and get -ENOENT, you can silently exit the state machine. Otherwise, you can continue to init and remove.
|
@ceph-jenkins retest this please |
|
@vshankar Passed a QA test suite run -- ready to merge if you tell me you are finished. |
Introduce static member function in ImageCtx to fetch (and initialize) singleton thread pool. Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Asynchronous removal of journal objects for a given rbd image. Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Introduce asynchronous journal object creation. This is a prerequisite for asynchronous creation of rbd image journals. Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
... plus changes in rbd class library to use helpers methods. Signed-off-by: Venky Shankar <vshankar@redhat.com>
create_v2() uses util::generate_image_id() therefore not requiring 'bid' to be passed in as parameter. This makes bid generation in create() unnecessary -- hence move this call to create_v1(). Signed-off-by: Venky Shankar <vshankar@redhat.com>
Use the newly instroduced asynchronous image creation state machine (CreateRequest) to create mirrored images. Signed-off-by: Venky Shankar <vshankar@redhat.com>
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Synchronous callers now call the asynchronous version wrapped around C_SaferCond. Also take care of mocked methods. Fixes: http://tracker.ceph.com/issues/15321 Signed-off-by: Venky Shankar <vshankar@redhat.com>
No description provided.