# Restarts the vmagent deployment; errors are intentionally ignored.
function metricRestart() {
    k3sRun rollout restart deployment/"$metricKube"-vmagent || true
}

# Writes Prometheus metrics (build info, vhost counts, domain counts) to $metricPromPath/kube.prom.
function metricKube() {
    local fileProm="$metricPromPath/kube.prom"
    local fileTmp="${fileProm}.tmp"

    mkdir -p -- "$metricPromPath" || return 1

    local vhostAllCount vhostUpCount
    vhostAllCount=$(openlitespeedVhostList all | grep -c . || true)
    vhostUpCount=$(openlitespeedVhostList up | grep -c . || true)

    local appTimestamp
    appTimestamp=$(date -d "$appDate ${appTime//-/:}" +%s)

    local domainAllCount=-1 domainFreeCount=-1
    [[ -f "$domainsList" ]] && domainAllCount=$(grep -cP '^(?!\s*$)' "$domainsList" || true)
    [[ -f "$domainsList" ]] && domainFreeCount=$(grep -c '^#' "$domainsList" || true)

    cat > "$fileTmp" <<EOF
# HELP kube_build_info Build and version info
# TYPE kube_build_info gauge
kube_build_info{version="${appVersion}"} 1

# HELP kube_start_time Time snapshot
# TYPE kube_start_time gauge
kube_start_time $appTimestamp

# HELP kube_vhost_all_count Number of virtual hosts
# TYPE kube_vhost_all_count gauge
kube_vhost_all_count $vhostAllCount

# HELP kube_vhost_up_count Number of active virtual hosts
# TYPE kube_vhost_up_count gauge
kube_vhost_up_count $vhostUpCount

# HELP kube_domain_all_count Number of domains
# TYPE kube_domain_all_count gauge
kube_domain_all_count $domainAllCount

# HELP kube_domain_use_count Number of use domains
# TYPE kube_domain_use_count gauge
kube_domain_use_count $domainFreeCount
EOF
    mv "$fileTmp" "$fileProm"
}

# Collects lsphp CPU/memory/worker metrics per domain across OLS pods and writes to $metricPromPath/lsphp.prom.
function metricLsphp() {
    local fileProm="$metricPromPath/lsphp.prom"
    local fileTmp="${fileProm}.tmp"

    mkdir -p -- "$metricPromPath" || return 1

    local podList error
    run podList error k3sPodListStatus "$openlitespeedKube" || return 1

    local pod phase output cpu mem count domain
    {
        printf '# HELP lsphp_cpu_percent Total CPU percent used by lsphp workers per domain\n'
        printf '# TYPE lsphp_cpu_percent gauge\n'
        printf '# HELP lsphp_memory_percent Total memory percent used by lsphp workers per domain\n'
        printf '# TYPE lsphp_memory_percent gauge\n'
        printf '# HELP lsphp_worker_count Number of lsphp worker processes per domain\n'
        printf '# TYPE lsphp_worker_count gauge\n'

        while IFS='|' read -r pod phase; do
            [[ "$phase" == "Running" ]] || continue
            run output error openlitespeedPodExec "$pod" sh -c "ps aux | grep lsphp | grep -v grep | awk '{u=\$1;cpu[u]+=\$3;mem[u]+=\$4;count[u]++} END {for(u in cpu){name=u;cmd=\"find /var/www/data -maxdepth 1 -uid \"u\" -type d 2>/dev/null\";if((cmd|getline dir)>0&&dir!=\"\"){n=split(dir,a,\"/\");name=a[n]};close(cmd);print cpu[u],mem[u],count[u],name}}'" || continue
            while IFS=' ' read -r cpu mem count domain; do
                [[ -n "$domain" ]] || continue
                printf 'lsphp_cpu_percent{domain="%s",pod="%s"} %s\n' "$domain" "$pod" "$cpu"
                printf 'lsphp_memory_percent{domain="%s",pod="%s"} %s\n' "$domain" "$pod" "$mem"
                printf 'lsphp_worker_count{domain="%s",pod="%s"} %s\n' "$domain" "$pod" "$count"
            done <<< "$output"
        done < <(awk 'NF' <<< "$podList")
    } > "$fileTmp"

    mv "$fileTmp" "$fileProm"
}

# Collects disk usage per site and sends metrics to the API.
function metricSites() {
    local metricSitesApiUrl="https://api.sodew.ai/api/metrics/sites"
    local batchId batchTime metricsJson dataJson payload httpCode
    batchId="$(uuidgen | tr '[:upper:]' '[:lower:]')"
    batchTime="$(date +%s)"
    metricsJson='{
    "site_disk_usage_mb":  {"type":"number"}
    }'
    dataJson='{}'

    local error vhostList
    run vhostList error openlitespeedVhostList all || {
        appError "Error retrieving vhost list: $error"
        return 1
    }
    while IFS= read -r domain <&3; do
        local diskUsage domainJson
        diskUsage="$(pathSize "$vhostsPath/$domain" "mb" || true)"
        [[ -n "$diskUsage" ]] || continue

        domainJson="$(
            jq -n --arg diskUsage "$diskUsage" \
                '{
                    site_disk_usage_mb: { value: (if ($diskUsage | length) > 0 then ($diskUsage | tonumber) else null end) },
                }'
        )"
        dataJson="$(jq --arg domain "$domain" --argjson row "$domainJson" '. + {($domain): $row}' <<< "$dataJson")"
    done 3< <(awk 'NF' <<< "$vhostList")

    payload="$(
        jq -n \
            --arg source_type "worker" \
            --arg source_id "$hostUuid" \
            --arg batch_id "$batchId" \
            --argjson batch_time "$batchTime" \
            --argjson metrics "$metricsJson" \
            --argjson data "$dataJson" \
            '{
                source_type: $source_type,
                source_id: $source_id,
                batch_id: $batch_id,
                batch_time: $batch_time,
                metrics: $metrics,
                data: $data
            }'
    )"

    # Sending data...
    httpCode="$(curl -sS -o /tmp/metrics_sites_response.txt -w '%{http_code}' -X POST "$metricSitesApiUrl" -H 'Content-Type: application/json' --data-binary "$payload")"
    if [[ "$httpCode" != "204" ]]; then
        appError "Ingest failed, HTTP $httpCode"
        cat /tmp/metrics_sites_response.txt >&2 || true
        return 1
    fi
}