Helper Commands for Kubernetes

load | stat | help | version

warm up cache

$ kubectl exec <pod-name> -- mapfs load <filename>

display volume detailed status

$ kubectl exec <pod-name> -- mapfs stat
============ Volume ============
VolumeName:     vol1
MountPoint:     /mnt/fuse

=== Memory Usage ===
Write Buffer:   0 Bytes
Read Buffer:    0 Bytes
ReadAhead Buffer:       0 Bytes
MetaCache:      144.00 MB

=== Requests ===
Total Request:  28

COMMAND         Request Total(ms)       Avg(ms) Min(ms) Max(ms)
lookup:         0       0       0       0       0
getattr:        3       17      5       0       17
readdir:        4       33      8       2       18
create:         0       0       0       0       0
unlink:         0       0       0       0       0
mkdir:          0       0       0       0       0
rmdir:          0       0       0       0       0
open:           1       0       0       0       0

=== IO Statistics ===
Total Read:     1.03 MB
Read Cloud:     1.16 MB
Write IO:       0 Bytes

=== Network Statistics ===
dns:            0 ms
connect:        0 ms
ssl:            0 ms
pretransfer:    1 ms
starttransfer:  41 ms
total:          45 ms
INDEX    CPU%

=== Cache Statistics ===
Data Cache Hits:        -
ReadAhead Hits:         87%
MetaData Query:         26
Meta Cache Hits:        100%

help & versison

$ kubectl exec <pod-name> -- mapfs help
$ kubectl exec <pod-name> -- mapfs version

Show ChannelServer connected clients

This one runs against the ChannelServer (channeld) itself -- the cache-invalidation fan-out server every mount-daemon pod connects to (see channel-server.yaml) -- not a per-volume mount-daemon pod like the commands above:

$ kubectl -n kube-system exec deploy/mapfs-channel -- channeld show
Channel                 Index   IP              Joined
a1b2c3d4e5f6a7b8         5       10.0.1.23       Yes
a1b2c3d4e5f6a7b8         7       10.0.2.45       Yes
c9d8e7f6a5b4c3d2         9       10.0.3.12       No

Each row is one mapfs client (a mount-daemon pod, somewhere in the cluster) currently exchanging cache-invalidation metadata through this ChannelServer:

• Channel -- hex fingerprint identifying the cache-invalidation group. Clients mounting the same volume share a fingerprint, so an invalidation fans out only within that group, not to every connected client.
• Index -- the client's internal socket index.
• IP -- the connecting mount-daemon pod's IP.
• Joined -- whether the client has completed its join handshake yet; briefly No right after it connects, before its first fingerprint announcement.

The total number of mapfs clients currently connected is just the row count below the header, e.g.:

$ kubectl -n kube-system exec deploy/mapfs-channel -- channeld show | tail -n +2 | wc -l

If nothing is connected, the command prints No clients connected instead of the table.