[Gluster-devel] AFR / Self Heal

[Gareth Bult]

Hi, 

At the moment, to kick in a self-heal on an AFR volume, it seems you need to 
close and re-open a given file. 

Also, while a self heal is in progress, the file will not open fully and cannot 
be used until the heal completes. 

Looking at the roadmap for 1.4 the implication is that these features may be 
added .. 

Can anyone can confirm this and possibly indicate how long it's likely to take 
for this to become available ? 

just fyi; 

Using XEN images at the moment on gluster FS it goes like this; 

Migration 
======= 

xm migrate -l domain nodeb -- takes 3-4 seconds with less than 1 second 
downtime 

Gluster restarts 
============ 

killall glusterfsd 
glusterfsd 
(all images are now running off other nodes) 

For each XEN instance; 

xm save domain /tmp/domain (4 seconds) 
head -c1 /gluster/images/domain.disk > /dev/null (~20 seconds) 
xm restore /tmp/domain (4 seconds) 


1) being able to start a self-heal and simply blocking an open file would save 
30% of the time required 
2) not having to copy a 2-3Gb image but instead use failed transactions would 
probably save 95%+ 

Gareth.