[WIP] lib, mgmtd: Fix the Scratch Buffer usage in Northbound code #14594
Conversation
|
Did you test this? Just quickly looked at the code and I see several scenarios where I think it won't work. Please confirm that you've tested it on various scenarios (create/modify/delete data, leaves/containers, with existing and non-existing parents, etc.), before I actually start to review this. It would be also great to see some numbers on performance difference between this and full diff. |
pushpasis
force-pushed
the
nb_scratchbuf_fix
branch
2 times, most recently
from
694d618 to
884fa22
Compare
@idryzhov I did test quiet a few combination of scenarios... like..
In each case the extra log added in mgmt_txn.c was used to verify that we are using the scratch buffer. Also ran the mgmt topotests before uploading the PR. But haven't run the full TopoTests yet. Give me some time to finish those. I will suggest you to start reviewing once all the topotests in CI runs without failure. |
pushpasis
force-pushed
the
nb_scratchbuf_fix
branch
from
884fa22 to
59ac868
Compare
The changes to the candidate data trees were not getting added to
the scratch-buffer since lyd_diff_get_op was returning nothing ('n')
as no 'operational' metadata associated with the data items in
the YANG models. Fixed the logic to add them to the scratch buffer
anyways.
Also, the scratch-buffer updations were getting invoked everywhere
(mgmtd, backend clients). Limited the usage to mgmtd daemon only.
Signed-off-by: Pushpasis Sarkar <[email protected]>
pushpasis
force-pushed
the
nb_scratchbuf_fix
branch
from
59ac868 to
e751574
Compare
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
pushpasis
changed the title
pushpasis
changed the title
|
This PR is stale because it has been open 180 days with no activity. Comment or remove the |
The changes to the candidate data trees were not getting added to the scratch-buffer since lyd_diff_get_op was returning nothing ('n') as no 'operational' metadata associated with the data items in the YANG models. Fixed the logic to add them to the scratch buffer anyways.
Also, the scratch-buffer updations were getting invoked everywhere (mgmtd, backend clients). Limited the usage to mgmtd daemon only.