summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGravatar karl <karl@web>2016-03-13 16:01:03 +0000
committerGravatar admin <admin@branchable.com>2016-03-13 16:01:03 +0000
commit69ceed50972f8f4a07b92e6c920496063a1d75ee (patch)
tree9083fda58484a180201778807c74d88095826525
parentd80fc9db8c98f02181a5a3de56e8e8e6b17e2dda (diff)
-rw-r--r--doc/todo/immediate_stream-to-sync.mdwn9
1 files changed, 9 insertions, 0 deletions
diff --git a/doc/todo/immediate_stream-to-sync.mdwn b/doc/todo/immediate_stream-to-sync.mdwn
new file mode 100644
index 000000000..919651df5
--- /dev/null
+++ b/doc/todo/immediate_stream-to-sync.mdwn
@@ -0,0 +1,9 @@
+git-annex is really great for secure distributed archival, but it's missing a piece of the reliability portion: when data is generated and saved a system may crash before finishing, and some data may be lost, before the add is initiated.
+
+If there were a way to add a file at the same time as writing it -- and then at the same time stream that live data to a remote -- it would increase the utility of git-add for such uses as storing live recording and logfiles.
+
+I imagine e.g.
+
+$ cat /proc/kmsg | git annex addstreamsync livelogs/kernel-log-$(date)
+
+and kmsg would be streamed to enough remotes to meet numcopies, and hashed live . If possible, if the system then crashed and even the entire source repo was lost, the remote would have some incomplete file and be able to complete the add properly at next sync. Hashing the data at the same time as generating it would also speed adding any new file.