Lucas Di Pentima [Tue, 11 Oct 2016 17:52:22 +0000 (14:52 -0300)]
9701: Wrapping actual alloc_bufferblock and delete_bufferblock methods inside synchronized calls, to be able to use them from within already synchronized BlockManager methods.
radhika [Tue, 11 Oct 2016 14:08:45 +0000 (10:08 -0400)]
10078: Rename the workbench config param 'show_user_notifications_on_dashboard' as 'show_user_notifications' because they are in topnav, not on dashboard!
radhika [Fri, 7 Oct 2016 17:40:34 +0000 (13:40 -0400)]
10078: Since we are not fetching components for pipeline instances (at least until the performance issues are resolved),
we will no longer see the running and queued jobs listed within each pipeline instance row in dashboard. Hence, update
test to look for these for a container request row instead of for a pipeline instance row.
radhika [Fri, 7 Oct 2016 14:19:48 +0000 (10:19 -0400)]
10078: When fetching recent_processes, do not select 'components' which is not needed in this context. For very large pipelines, this will perform better.
radhika [Mon, 3 Oct 2016 22:58:59 +0000 (18:58 -0400)]
10078: Now that the recent_collections panel includes collections from Home directory as well
(self and other users' for admin user), suppress displaying confusing '(none)' in place of a Home directory.
10078: Update recent_collections so that the collection query it is not executed twice, once in this method and once in show_dashboard for r[:collections].each statement.
10078: Update the Node.filter so that this query is not executed three times: once for nodes.sort_by and twice in compute_node_summary for nodes.select
radhika [Fri, 7 Oct 2016 17:40:34 +0000 (13:40 -0400)]
10078: Since we are not fetching components for pipeline instances (at least until the performance issues are resolved),
we will no longer see the running and queued jobs listed within each pipeline instance row in dashboard. Hence, update
test to look for these for a container request row instead of for a pipeline instance row.
radhika [Fri, 7 Oct 2016 14:19:48 +0000 (10:19 -0400)]
10078: When fetching recent_processes, do not select 'components' which is not needed in this context. For very large pipelines, this will perform better.
Lucas Di Pentima [Thu, 6 Oct 2016 20:23:09 +0000 (17:23 -0300)]
9701: Several corrections/enhancements:
* Added a 'sync' parameter on repack_small_blocks() so that blocks commits can
be done in either way depending from where it's called.
* Allow packing small buffer blocks up to a full block on Keep.
* Replaced ArvadosFile's _closed flag with a list of its ArvadosFileWriter objects,
so that it's used as a reference counter.
* Moved ArvadosFile flush behaviour from ArvadosFileWriter.close() method to
ArvadosFile.remove_writer() so that it can decide whether it should commit the
buffer block or repack it with others.
Lucas Di Pentima [Wed, 5 Oct 2016 18:20:55 +0000 (15:20 -0300)]
9701: Changes on the Python SDK to allow small file packing on Collection class:
* Added optional flush parameter to ArvadosFileWriter.close().
* Added _closed attribute & related accessors to ArvadosFile to enable BlockManager to query this state.
* For every ArvadosFile close() operation, call BlockManager to search for small blocks for repacking purposes.
* Do a last repacking operation just before exiting, joining the last small blocks even if they don't meet the
minimum size requirement.
There's some pending code cleanup, as BlockManager fiddles directly with ArvadosFile._segments lists, some locking:
issues have to be solved.
radhika [Mon, 3 Oct 2016 22:58:59 +0000 (18:58 -0400)]
10078: Now that the recent_collections panel includes collections from Home directory as well
(self and other users' for admin user), suppress displaying confusing '(none)' in place of a Home directory.
Peter Amstutz [Fri, 30 Sep 2016 19:59:51 +0000 (15:59 -0400)]
10165: Add FinalOutputPathMapper. Delete basename/size/listing from output object because that information is captured by the enclosing Collection. Sort keys.
10078: Update recent_collections so that the collection query it is not executed twice, once in this method and once in show_dashboard for r[:collections].each statement.
10078: Update the Node.filter so that this query is not executed three times: once for nodes.sort_by and twice in compute_node_summary for nodes.select