Description of problem:

During the introduction of snapshot, in glusterfs-3.6, a change was done in the way the /var/lib/glusterfs/vols/<vol>/info file was read, and a new limitation of 4096 (PATH_MAX) character per line was introduce.

- Patch introducing the limitation : http://review.gluster.org/#/c/7128/
- in particular : http://review.gluster.org/#/c/7128/21/libglusterfs/src/store.c , line 182

Version-Release number of selected component (if applicable): glusterfs-3.6 and above


How reproducible: 100%


Steps to Reproduce:
1. set a key/value longer than 4096 characters for a volume
2. restart glusterd

Actual results:
fail


Expected results:
no fail, or at the very least a better explanation

Additional info:

This limitation was introduced with snapshots : http://review.gluster.org/#/c/7128/

before, there was no limitation :
----8<---
int
gf_store_read_and_tokenize (FILE *file, char *str, char **iter_key,
                            char **iter_val, gf_store_op_errno_t *store_errno)
{
[...]

        ret = fscanf (file, "%s", str);
        if (ret <= 0 || feof (file)) {
                ret = -1;
                *store_errno = GD_STORE_EOF;
                goto out;
        }
--->8---

after, the max length is PATH_MAX, which is 4096 :
---8<---
int
gf_store_read_and_tokenize (FILE *file, char *str, char **iter_key,
                            char **iter_val, gf_store_op_errno_t *store_errno)
{
[...]
        temp = fgets (str, PATH_MAX, file);
        if (temp == NULL || feof (file)) {
                ret = -1;
                *store_errno = GD_STORE_EOF;
                goto out;
        }

[...]
--->8---


Quick reproducer :

[root@cbuissar-rhel7-rhs31-node1 rpms]# LIST=$(for IP in 122.{0..255} ; do echo -n "192.168.$IP,"  ; done ; for IP in 123.{0..6} ; do echo -n "192.168.$IP" ; [[ $IP =~ 123.6 ]] || echo -n ","  ; done ;)
[root@cbuissar-rhel7-rhs31-node1 rpms]# echo auth.allow=$LIST | wc -c
4095
[root@cbuissar-rhel7-rhs31-node1 rpms]# gluster volume set slim auth.allow $LIST
volume set: success
[root@cbuissar-rhel7-rhs31-node1 rpms]# systemctl restart glusterd.service 
[root@cbuissar-rhel7-rhs31-node1 rpms]# echo $?
0
[root@cbuissar-rhel7-rhs31-node1 rpms]# LIST="$LIST,192.168.123.7"
[root@cbuissar-rhel7-rhs31-node1 rpms]# echo auth.allow=$LIST | wc -c
4109
[root@cbuissar-rhel7-rhs31-node1 rpms]# gluster volume set slim auth.allow $LIST
volume set: success
[root@cbuissar-rhel7-rhs31-node1 rpms]# systemctl restart glusterd.service 
Job for glusterd.service failed. See 'systemctl status glusterd.service' and 'journalctl -xn' for details.

will attach the logs.

=> This will be considered as regression by customer upgrading from RHGS 2.1, which did not seem to have this limitation : glusterd refuses to start after upgrade to 3.X

=> If we need to keep a limitation, we should probably at least expand to a bigger value than 4096.
=> However, at the minimum, we should test the length, and if we reached the limit, either truncate/cancel, or fail all together, but with a clear error (you can't guess the issue unless you are in DEBUG level)