0.5-64

hive/bin/agent

@@ -32,6 +32,9 @@ gpu_stats=
 #nvidia_indexes=echo "$gpu_stats" | jq '.brand | to_entries[] | select(.value == "nvidia") | .key' | jq -sc '.'
 #nvidia_indexes=`gpu-detect listjson | jq '. | to_entries[] | select(.value.brand == "nvidia") | .key' | jq -sc '.'`
 #gpu_detect_json=`gpu-detect listjson`
+if [[ ! -e /run/hive/gpu-detect.json ]]; then #this is needed only when upgrading version. later we can remove this
+	gpu-detect listjson > /run/hive/gpu-detect.json
+fi
 gpu_detect_json=`cat /run/hive/gpu-detect.json`
 amd_indexes_array=`echo "$gpu_detect_json" | jq -c '[ . | to_entries[] | select(.value.brand == "amd") | .key ]'`
 nvidia_indexes_array=`echo "$gpu_detect_json" | jq -c '[ . | to_entries[] | select(.value.brand == "nvidia") | .key ]'`
@@ -48,10 +51,8 @@ cpu_indexes_array=`echo "$gpu_detect_json" | jq -c '[ . | to_entries[] | select(
 function reset_miner_vars() {
 	#reset all miner vars
 	MINER=
-	COIN=
 	for i in {1..32}; do
 		eval unset MINER$i
-		eval unset COIN$i
 	done
 }
 
@@ -89,16 +90,14 @@ function loop {
 	local mindex
 	for mindex in {32..2}; do
 		eval minerval=\$MINER$mindex
-		eval coinval=\$COIN$mindex
 		if [[ ! -z $minerval ]]; then
 			miner_stats $minerval $mindex
 			eval khs$mindex=\$khs
 			miner_json=$(jq -n -c \
 				--arg miner$mindex "$minerval" \
-				--arg coin$mindex "$coinval" \
 				--arg total_khs$mindex "$khs" \
 				--argjson miner_stats$mindex "$stats" \
-				'{"params": {$miner'$mindex', $coin'$mindex', $total_khs'$mindex', $miner_stats'$mindex'}}')
+				'{"params": {$miner'$mindex', $total_khs'$mindex', $miner_stats'$mindex'}}')
 			eval miner${mindex}_json="\$miner_json"
 		fi
 	done
@@ -125,24 +124,26 @@ function loop {
 	#echo $stats
 	#uptime: load averages for the past 1, 5, and 15 minutes
 
+	[[ ! -z $META ]] && meta="$META" || meta='null'
 	request=$(
 	jq -n \
 	--arg rig_id "$RIG_ID" \
 	--arg passwd "$RIG_PASSWD" \
 	--arg miner "$MINER" \
-	--arg coin "$COIN" \
+	--argjson meta "$meta" \
 	--argjson miner_stats "$stats" \
 	--arg total_khs "$khs" \
 	--argjson temp "$temp" \
 	--argjson fan "$fan" \
 	--argjson power "$power" \
 	--arg df "`df -h / | awk '{ print $4 }' | tail -n 1 | sed 's/%//'`" \
+	--argjson mem "`free -m | grep 'Mem' | awk '{print "["$2","$4"]"}'`" \
 	--argjson cpuavg "[`uptime | awk -F': ' '{print $2}'`]" \
 	'{
 		"method": "stats", "jsonrpc": "2.0", "id": 0,
 		"params": {
-			$rig_id, $passwd, $miner, $coin, $miner_stats, $total_khs,
-			$temp, $fan, $power, $df, $cpuavg
+			$rig_id, $passwd, $miner, $meta, $miner_stats, $total_khs,
+			$temp, $fan, $power, $df, $mem, $cpuavg
 		}
 	}'
 	)

hive/bin/agent.miner_stats.sh

@@ -193,12 +193,13 @@ function miner_stats {
 				khs=`echo $stats_raw | jq '.result[].sol_ps' | awk '{s+=$1} END {print s/1000}'`
 				local uptime=$(( `date +%s` - $(stat -c%X /proc/`pidof zm`) )) #dont think zm will die so soon after getting stats
 #				local fan=$(jq -c "[.fan$nvidia_indexes_array]" <<< $gpu_stats)
-				local temp=$(jq -c "[.temp$nvidia_indexes_array]" <<< $gpu_stats)
+#				local temp=$(jq -c "[.temp$nvidia_indexes_array]" <<< $gpu_stats)
+				local temp=$(jq -c '[.result[].temperature]' <<< "$stats_raw")
 				local ac=`echo $stats_raw | jq '[.result[].accepted_shares] | add'`
 				local rj=`echo $stats_raw | jq '[.result[].rejected_shares] | add'`
 
 				#All fans speed array
-				local fan=$(jq -r ".fan | .[]" <<< $gpu_stats)
+				local fan=$(jq -r ".fan | .[]" <<< "$gpu_stats")
 				#DSTM's busid array
 				local bus_id_array=$(jq -r '.result[].gpu_pci_bus_id' <<< "$stats_raw")
 				#All busid array
@@ -219,7 +220,8 @@ function miner_stats {
 					done
 				done
 
-				stats=$(jq --argjson temp "$temp" --argjson fan "`echo "${fans_array[@]}" | jq -s . | jq -c .`" --arg uptime "$uptime" --arg ac "$ac" --arg rj "$rj" \
+				stats=$(jq --argjson temp "$temp" --argjson fan "`echo "${fans_array[@]}" | jq -s . | jq -c .`" \
+					--arg uptime "$uptime" --arg ac "$ac" --arg rj "$rj" \
 					'{ hs: [.result[].sol_ps], $temp, $fan, $uptime, ar: [$ac, $rj] }' <<< "$stats_raw")
 			fi
 		;;

hive/bin/hello

@@ -83,10 +83,11 @@ jq -n \
 --arg manufacturer "$mb_manufacturer" --arg product "$mb_product" \
 --arg model "$cpu_model" --arg cores "$cpu_cores" --arg aes "$aes" \
 --arg kernel "$kernel" \
+--arg server_url "$HIVE_HOST_URL" \
 '{
 	"method": "hello", "jsonrpc": "2.0", "id": 0,
 	"params": {
-		$rig_id, $passwd, $farm_hash,
+		$rig_id, $passwd, $farm_hash, $server_url,
 		$boot_time, $boot_event, $ip, $net_interfaces, $openvpn,
 		$gpu, $gpu_count_amd, $gpu_count_nvidia,
 		"mb": {$manufacturer, $product}, "cpu": {$model, $cores, $aes},

hive/sbin/autofan

@@ -38,6 +38,12 @@ DEF_REBOOT_ON_ERROR=0
 
 #we will restart miner only it was stopped by this script
 miner_stopped_by_overheat=0
+#flag that the message was sent
+unable_to_set_fan_speed=0
+#flag if we met 511, 460, etc
+temperature_is_unreal=0
+#unparsable data
+error_in_temp_readings=0
 
 ###
 # Log write
@@ -126,12 +132,19 @@ get_fan_speed () {
 	#checking var is natural number
 	if [[ -z "${cur_temp##*[!0-9]*}" || -z "${cur_fan_speed##*[!0-9]*}" ]]; then
 		if [[ $REBOOT_ON_ERROR == 1 ]]; then
-			local msg="Autofan: GPU error, rebooting"
+			local msg="Autofan: error in temp readings, rebooting"
 			local payload=`cat $gpu_stats_json | jq -r -C .`
 			echo "$payload" | message warning "$msg" payload
 			nohup bash -c 'sreboot' > /tmp/nohup.log 2>&1 &
-		else break
+		else
+			if [[ $error_in_temp_readings == 0 ]]; then
+				local msg="Autofan: error in temp readings"
+				local payload=`cat $gpu_stats_json | jq -r -C .`
+				echo "$payload" | message warning "$msg" payload
+				error_in_temp_readings=1
+			fi
 		fi
+		break
 	fi
 	#
 
@@ -248,21 +261,30 @@ check_overheat() {
 		local t
 		for t in ${temperatures_array[@]}; do
 			#reboot on driver error
-			if [[ $REBOOT_ON_ERROR == 1 && $t -gt 120 ]]; then
-				local msg="GPU temperature $t is unreal, driver error"
-				local payload=`cat $gpu_stats_json | jq -r -C .`
-				echo "$payload" | message warning "$msg"
-				nohup bash -c 'sreboot' > /tmp/nohup.log 2>&1 &
+			if [[ $t -gt 120 ]]; then
+				if [[ $REBOOT_ON_ERROR == 1 ]]; then
+					local msg="GPU temperature $t is unreal, driver error, rebooting"
+					local payload=`cat $gpu_stats_json | jq -r -C .`
+					echo "$payload" | message warning "$msg" payload
+					nohup bash -c 'sreboot' > /tmp/nohup.log 2>&1 &
+				else
+					if [[ $temperature_is_unreal == 0 ]]; then
+						local msg="GPU temperature $t is unreal, driver error"
+						local payload=`cat $gpu_stats_json | jq -r -C .`
+						echo "$payload" | message warning "$msg" payload
+						temperature_is_unreal=1
+					fi
+				fi
 				break
 			fi
 
 			#stop on CRITICAL_TEMP
-			if [[ $t -ge $CRITICAL_TEMP && $t -ne 511 ]]; then
+			if [[ $t -ge $CRITICAL_TEMP && $t -le 120 ]]; then #do not process temp 511, 460, etc
 				miner_stopped_by_overheat=1
 				do_miner_action "miner-stop"
 				local msg="GPU reached ${CRITICAL_TEMP}°C, mining stopped"
 				local payload=`cat $gpu_stats_json | jq -r -C .`
-				echo "$payload" | message warning "$msg"
+				echo "$payload" | message warning "$msg" payload
 				break
 			fi
 		done
@@ -302,8 +324,7 @@ nvidia_auto_fan_control () {
 	args=
 	#set start index
 	i=0
-	for index in ${nvidia_indexes_array[@]}
-	do
+	for index in ${nvidia_indexes_array[@]}; do
 		# TODO Theese fields maybe moved inside `get_fan_speed` replaced by on nvidia_indexes_array[@] as argument
 		local gpu_temperature=${temperatures_array[index]}
 		local gpu_temperature_previous=${temperatures_array_previous[index]}
@@ -321,14 +342,18 @@ nvidia_auto_fan_control () {
 	#[[ -n $args ]] && nvidia-settings $args > /dev/null 2>&1
 	if [[ -n $args ]]; then
 		timeout 30 nvidia-settings $args > /dev/null 2>&1
-		if [[ $? -ne 0 && $REBOOT_ON_ERROR == 1 ]]; then
-			local msg="Autofan: unable to set fan speed, rebooting"
-			message warning "$msg"
-			nohup bash -c 'sreboot' > /tmp/nohup.log 2>&1 &
-		fi
-		if [[ $? -ne 0 && $REBOOT_ON_ERROR == 0 ]]; then
-			local msg="Autofan: unable to set fan speed"
-			message warning "$msg"
+		if [[ $? -ne 0 ]]; then
+			if [[ $REBOOT_ON_ERROR == 1 ]]; then
+				local msg="Autofan: unable to set fan speed, rebooting"
+				message warning "$msg"
+				nohup bash -c 'sreboot' > /tmp/nohup.log 2>&1 &
+			else
+				if [[ $unable_to_set_fan_speed == 0 ]]; then
+					local msg="Autofan: unable to set fan speed"
+					message warning "$msg"
+				fi
+			fi
+			unable_to_set_fan_speed=1
 		fi
 	fi
 }