A couple of days ago, I had some issues with NFS consistency, not every servers were up to date. Some servers had the good version of the file some hadn’t. However performing a ls -l seemed to fixed tempory the problem after each update (where a simple ls didn’t). Indeed issuing ls with the -l option triggers stat() where ls doesn’t, just because file attributes are called from the stats() function. I needed to investigate…
We determined that some web virtual didn’t delivered the same content, some page weren’t updated properly. We quickly figured out that the issue was located of the NFS attribute caching, by default too long, at least in our setup. The first change we made was to enable the noac option on the client mount. However while trying to enhance the consistency of the NFS data, the performance impact was pretty high. The impact was easy to detect and reproduce. Basically everytime a page was request from the webser, the client had to request the NFS server to be sure to del