summaryrefslogtreecommitdiff
path: root/doc/todo/parallel_get.mdwn
diff options
context:
space:
mode:
Diffstat (limited to 'doc/todo/parallel_get.mdwn')
-rw-r--r--doc/todo/parallel_get.mdwn85
1 files changed, 85 insertions, 0 deletions
diff --git a/doc/todo/parallel_get.mdwn b/doc/todo/parallel_get.mdwn
new file mode 100644
index 000000000..fb3f8d098
--- /dev/null
+++ b/doc/todo/parallel_get.mdwn
@@ -0,0 +1,85 @@
+Wish: `git annex get [files] -jN` should run up to N downloads of files
+concurrently.
+
+This can already be done by just starting up N separate git-annex
+processes all trying to get the same files. They'll coordinate themselves
+to avoid downloading the same file twice.
+
+But, the output of concurrent git annex get's in a single teminal is a
+mess.
+
+It would be nice to have something similar to docker's output when fetching
+layers of an image. Something like:
+
+ get foo1 ok
+ get foo2 ok
+ get foo3 -> 5% 100 KiB/s
+ get foo4 -> 3% 90 KiB/s
+ get foo5 -> 20% 1 MiB/s
+
+Where the bottom N lines are progress displays for the downloads that are
+currently in progress. When a download finishes, it can scroll up the
+screen with "ok".
+
+ get foo1 ok
+ get foo2 ok
+ get foo5 ok
+ get foo3 -> 5% 100 KiB/s
+ get foo4 -> 3% 90 KiB/s
+ get foo6 -> 0% 110 Kib/S
+
+This display could perhaps be generalized for other concurrent actions.
+For example, drop:
+
+ drop foo1 ok
+ drop foo2 failed
+ Not enough copies ...
+ drop foo3 -> (checking r1...)
+ drop foo4 -> (checking r2...)
+
+But, do get first.
+
+Pain points:
+
+1. Currently, git-annex lets tools like rsync and wget display their own
+ progress. This makes sense for the single-file at a time get, because
+ rsync can display better output than just a percentage. (This is especially
+ the case with aria2c for torrents, which displays seeder/leecher info in
+ addition to percentage.)
+
+ But in multi-get mode, the progress display would be simplified. git-annex
+ can already get percent done information, either as reported by individiual
+ backends, or by falling back to polling the file as it's downloaded.
+
+2. The mechanics of updating the screen for a multi-line progress output
+ require some terminal handling code. Using eg, curses, in a mode that
+ doesn't take over the whole screen display, but just moves the cursor
+ up to the line for the progress that needs updating and redraws that
+ line. Doing this portably is probably going to be a pain, especially
+ I have no idea if it can be done on Windows.
+
+ An alternative would be a display more like apt uses for concurrent
+ downloads, all on one line:
+
+ get foo1 ok
+ get foo2 ok
+ get [foo3 -> 5% 100 KiB/s] [foo4 -> 3% 90 KiB/s] [foo5 -> 20% 1 MiB/s]
+
+ The problem with that is it has to avoid scrolling off the right
+ side, so it probably has to truncate the line. Since filenames
+ are often longer than "fooN", it probably has to elipsise the filename.
+ This approach is just not as flexible or nice in general.
+
+See also: [[parallel_possibilities]]
+
+> I am looking at using the ascii-progress library for this.
+> It has nice support for multiple progress bars, and is portable.
+> I have filed 7 issues on it, around 4 of which need to get fixed before
+> it's suitable for git-annex to use.. --[[Joey]]
+
+>> `git annex get -JN` works now, but lacks any progress display.
+>> Waiting on some updates to ascii-progress. --[[Joey]]
+
+>>> Wrote concurrent-output; [[done]] --[[Joey]]
+
+[[!meta author=yoh]]