Zabbix introduces blocking to a non-blocking check

[AquaNewt]

I've written two scripts, one for discovering nfses specifically, and a second script to check if the nfs is actually working or if the nfs is hanging because either the server has vanished or there's a firewall blocking the connection.

discovery:

#!/bin/bash

mounts=$(mount -t nfs,nfs3,nfs4 | cut -d' ' -f3)

COMMA=''
echo '['

if [[ -n ${mounts} ]]; then
  while read -r line; do
    echo "${COMMA} { \"{#NFSMOUNT}\":\"${line}\" }"
    COMMA=','
  done <<<"$mounts"
fi

echo ']'
​

check:

#!/bin/bash

mount="$*"

read -r -t1 < <(stat -t "${mount}") || lsof -b 2>/dev/null | grep "${mount}"
echo $?
​

If everything works, we get 0, if it doesn't, 1 as stdout. When I block nfs in the filesystem and run the check manually in almost every conceivable way, the check-script runs in about a second and returns 1, as it should. I can run it as root, I can run it as zabbix with sudo (for the stat), no problems.

However, if the nfs is broken and the script is run either normally via zabbix or via `zabbix_agent2 -t nfs.check["/mnt/broken"]`, the check hangs for about a minute before being killed. The logs also show `failed to kill [sudo /etc/zabbix/userparams-scripts/nfs_check.sh "/mnt/broken"]: operation not permitted`. At absolutely no point should the script itself hang, nor should zabbix have a need of killing it because of a timeout.

I'm at my wit's end, is zabbix broken, or is there anything I can do on my end to have this check work?

[AquaNewt]

I've found the reason for this behaviour: `read -t<number>` apparently doesn't work when bash is noninteractive, it just hangs endlessly, the same can be achieved by running the check with `watch ./check.sh /mnt/broken`. The solution is to replace the read part with `timeout 1 stat -t "${mount}" >/dev/null`.