local comms = require("scada-common.comms") local const = require("scada-common.constants") local databus = require("reactor-plc.databus") local log = require("scada-common.log") local ppm = require("scada-common.ppm") local rsio = require("scada-common.rsio") local types = require("scada-common.types") local util = require("scada-common.util") local plc = {} local RPS_TRIP_CAUSE = types.RPS_TRIP_CAUSE local PROTOCOL = comms.PROTOCOL local DEVICE_TYPE = comms.DEVICE_TYPE local ESTABLISH_ACK = comms.ESTABLISH_ACK local RPLC_TYPE = comms.RPLC_TYPE local SCADA_MGMT_TYPE = comms.SCADA_MGMT_TYPE local AUTO_ACK = comms.PLC_AUTO_ACK local RPS_LIMITS = const.RPS_LIMITS -- I sure hope the devs don't change this error message, not that it would have safety implications -- I wish they didn't change it to be like this local PCALL_SCRAM_MSG = "pcall: Scram requires the reactor to be active." local PCALL_START_MSG = "pcall: Reactor is already active." -- RPS: Reactor Protection System
-- identifies dangerous states and SCRAMs reactor if warranted
-- autonomous from main SCADA supervisor/coordinator control ---@nodiscard ---@param reactor table ---@param is_formed boolean ---@param emer_cool nil|table emergency coolant configuration function plc.rps_init(reactor, is_formed, emer_cool) local state_keys = { high_dmg = 1, high_temp = 2, low_coolant = 3, ex_waste = 4, ex_hcoolant = 5, no_fuel = 6, fault = 7, timeout = 8, manual = 9, automatic = 10, sys_fail = 11, force_disabled = 12 } local self = { state = { false, false, false, false, false, false, false, false, false, false, false, false }, reactor_enabled = false, enabled_at = 0, emer_cool_active = nil, ---@type boolean formed = is_formed, force_disabled = false, tripped = false, trip_cause = "ok" ---@type rps_trip_cause } -- PRIVATE FUNCTIONS -- -- set reactor access fault flag local function _set_fault() if reactor.__p_last_fault() ~= "Terminated" then self.state[state_keys.fault] = true end end -- set emergency coolant control (if configured) ---@param state boolean true to enable emergency coolant, false to disable local function _set_emer_cool(state) -- check if this was configured: if it's a table, fields have already been validated. if type(emer_cool) == "table" then local level = rsio.digital_write_active(rsio.IO.U_EMER_COOL, state) if level ~= false then if rsio.is_color(emer_cool.color) then local output = rs.getBundledOutput(emer_cool.side) if rsio.digital_write(level) then output = colors.combine(output, emer_cool.color) else output = colors.subtract(output, emer_cool.color) end rs.setBundledOutput(emer_cool.side, output) else rs.setOutput(emer_cool.side, rsio.digital_write(level)) end if state ~= self.emer_cool_active then if state then log.info("RPS: emergency coolant valve OPENED") else log.info("RPS: emergency coolant valve CLOSED") end self.emer_cool_active = state end end end end -- check if the reactor is formed local function _is_formed() local formed = reactor.isFormed() if formed == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() else self.formed = formed if not self.state[state_keys.sys_fail] then self.state[state_keys.sys_fail] = not formed end end end -- check if the reactor is force disabled local function _is_force_disabled() local disabled = reactor.isForceDisabled() if disabled == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() else self.force_disabled = disabled if not self.state[state_keys.force_disabled] then self.state[state_keys.force_disabled] = disabled end end end -- check for high damage local function _high_damage() local damage_percent = reactor.getDamagePercent() if damage_percent == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() elseif not self.state[state_keys.high_dmg] then self.state[state_keys.high_dmg] = damage_percent >= RPS_LIMITS.MAX_DAMAGE_PERCENT end end -- check if the reactor is at a critically high temperature local function _high_temp() -- mekanism: MAX_DAMAGE_TEMPERATURE = 1_200 local temp = reactor.getTemperature() if temp == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() elseif not self.state[state_keys.high_temp] then self.state[state_keys.high_temp] = temp >= RPS_LIMITS.MAX_DAMAGE_TEMPERATURE end end -- check if there is very low coolant local function _low_coolant() local coolant_filled = reactor.getCoolantFilledPercentage() if coolant_filled == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() elseif not self.state[state_keys.low_coolant] then self.state[state_keys.low_coolant] = coolant_filled < RPS_LIMITS.MIN_COOLANT_FILL end end -- check for excess waste (>80% filled) local function _excess_waste() local w_filled = reactor.getWasteFilledPercentage() if w_filled == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() elseif not self.state[state_keys.ex_waste] then self.state[state_keys.ex_waste] = w_filled > RPS_LIMITS.MAX_WASTE_FILL end end -- check for heated coolant backup (>95% filled) local function _excess_heated_coolant() local hc_filled = reactor.getHeatedCoolantFilledPercentage() if hc_filled == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() elseif not self.state[state_keys.ex_hcoolant] then self.state[state_keys.ex_hcoolant] = hc_filled > RPS_LIMITS.MAX_HEATED_COLLANT_FILL end end -- check if there is no fuel local function _insufficient_fuel() local fuel = reactor.getFuelFilledPercentage() if fuel == ppm.ACCESS_FAULT then -- lost the peripheral or terminated, handled later _set_fault() elseif not self.state[state_keys.no_fuel] then self.state[state_keys.no_fuel] = fuel <= RPS_LIMITS.NO_FUEL_FILL end end -- PUBLIC FUNCTIONS -- ---@class rps local public = {} -- re-link a reactor after a peripheral re-connect ---@param new_reactor table reconnected reactor function public.reconnect_reactor(new_reactor) reactor = new_reactor end -- trip for lost peripheral function public.trip_fault() _set_fault() end -- trip for a PLC comms timeout function public.trip_timeout() self.state[state_keys.timeout] = true end -- manually SCRAM the reactor function public.trip_manual() self.state[state_keys.manual] = true end -- automatic SCRAM commanded by supervisor function public.trip_auto() self.state[state_keys.automatic] = true end -- trip for unformed reactor function public.trip_sys_fail() self.state[state_keys.fault] = true self.state[state_keys.sys_fail] = true end -- SCRAM the reactor now (blocks waiting for server tick) ---@return boolean success function public.scram() log.info("RPS: reactor SCRAM") reactor.scram() if reactor.__p_is_faulted() and (reactor.__p_last_fault() ~= PCALL_SCRAM_MSG) then log.error("RPS: failed reactor SCRAM") return false else self.reactor_enabled = false self.last_runtime = util.time_ms() - self.enabled_at return true end end -- start the reactor now (blocks waiting for server tick) ---@return boolean success function public.activate() if not self.tripped then log.info("RPS: reactor start") reactor.activate() if reactor.__p_is_faulted() and (reactor.__p_last_fault() ~= PCALL_START_MSG) then log.error("RPS: failed reactor start") else self.reactor_enabled = true self.enabled_at = util.time_ms() return true end else log.debug(util.c("RPS: failed start, RPS tripped: ", self.trip_cause)) end return false end -- automatic control activate/re-activate ---@return boolean success function public.auto_activate() -- clear automatic SCRAM if it was the cause if self.tripped and self.trip_cause == "automatic" then self.state[state_keys.automatic] = true self.trip_cause = RPS_TRIP_CAUSE.OK self.tripped = false log.debug("RPS: cleared automatic SCRAM for re-activation") end return public.activate() end -- check all safety conditions ---@nodiscard ---@return boolean tripped, rps_trip_cause trip_status, boolean first_trip function public.check() local status = RPS_TRIP_CAUSE.OK local was_tripped = self.tripped local first_trip = false if self.formed then -- update state parallel.waitForAll( _is_formed, _is_force_disabled, _high_damage, _high_temp, _low_coolant, _excess_waste, _excess_heated_coolant, _insufficient_fuel ) else -- check to see if its now formed _is_formed() end -- check system states in order of severity if self.tripped then status = self.trip_cause elseif self.state[state_keys.sys_fail] then log.warning("RPS: system failure, reactor not formed") status = RPS_TRIP_CAUSE.SYS_FAIL elseif self.state[state_keys.force_disabled] then log.warning("RPS: reactor was force disabled") status = RPS_TRIP_CAUSE.FORCE_DISABLED elseif self.state[state_keys.high_dmg] then log.warning("RPS: high damage") status = RPS_TRIP_CAUSE.HIGH_DMG elseif self.state[state_keys.high_temp] then log.warning("RPS: high temperature") status = RPS_TRIP_CAUSE.HIGH_TEMP elseif self.state[state_keys.low_coolant] then log.warning("RPS: low coolant") status = RPS_TRIP_CAUSE.LOW_COOLANT elseif self.state[state_keys.ex_waste] then log.warning("RPS: full waste") status = RPS_TRIP_CAUSE.EX_WASTE elseif self.state[state_keys.ex_hcoolant] then log.warning("RPS: heated coolant backup") status = RPS_TRIP_CAUSE.EX_HCOOLANT elseif self.state[state_keys.no_fuel] then log.warning("RPS: no fuel") status = RPS_TRIP_CAUSE.NO_FUEL elseif self.state[state_keys.fault] then log.warning("RPS: reactor access fault") status = RPS_TRIP_CAUSE.FAULT elseif self.state[state_keys.timeout] then log.warning("RPS: supervisor connection timeout") status = RPS_TRIP_CAUSE.TIMEOUT elseif self.state[state_keys.manual] then log.warning("RPS: manual SCRAM requested") status = RPS_TRIP_CAUSE.MANUAL elseif self.state[state_keys.automatic] then log.warning("RPS: automatic SCRAM requested") status = RPS_TRIP_CAUSE.AUTOMATIC else self.tripped = false self.trip_cause = RPS_TRIP_CAUSE.OK end -- if a new trip occured... if (not was_tripped) and (status ~= RPS_TRIP_CAUSE.OK) then first_trip = true self.tripped = true self.trip_cause = status -- in the case that the reactor is detected to be active, -- it will be scrammed shortly after this in the main RPS loop if we don't here if self.formed then if not self.force_disabled then public.scram() else log.warning("RPS: skipping SCRAM due to reactor being force disabled") end else log.warning("RPS: skipping SCRAM due to not being formed") end end -- update emergency coolant control if configured _set_emer_cool(self.state[state_keys.low_coolant]) -- report RPS status databus.tx_rps(self.tripped, self.state, self.emer_cool_active) return self.tripped, status, first_trip end ---@nodiscard function public.status() return self.state end ---@nodiscard function public.is_tripped() return self.tripped end ---@nodiscard function public.get_trip_cause() return self.trip_cause end ---@nodiscard function public.is_low_coolant() return self.states[state_keys.low_coolant] end ---@nodiscard function public.is_active() return self.reactor_enabled end ---@nodiscard function public.is_formed() return self.formed end ---@nodiscard function public.is_force_disabled() return self.force_disabled end -- get the runtime of the reactor if active, or the last runtime if disabled ---@nodiscard ---@return integer runtime time since last enable function public.get_runtime() return util.trinary(self.reactor_enabled, util.time_ms() - self.enabled_at, self.last_runtime) end -- reset the RPS ---@param quiet? boolean true to suppress the info log message function public.reset(quiet) self.tripped = false self.trip_cause = RPS_TRIP_CAUSE.OK for i = 1, #self.state do self.state[i] = false end if not quiet then log.info("RPS: reset") end end -- reset the automatic and timeout trip flags, then clear trip if that was the trip cause function public.auto_reset() self.state[state_keys.automatic] = false self.state[state_keys.timeout] = false if self.trip_cause == RPS_TRIP_CAUSE.AUTOMATIC or self.trip_cause == RPS_TRIP_CAUSE.TIMEOUT then self.trip_cause = RPS_TRIP_CAUSE.OK self.tripped = false log.info("RPS: auto reset") end end -- link functions with databus databus.link_rps(public.trip_manual, public.reset) return public end -- Reactor PLC Communications ---@nodiscard ---@param id integer reactor ID ---@param version string PLC version ---@param modem table modem device ---@param local_port integer local listening port ---@param server_port integer remote server port ---@param range integer trusted device connection range ---@param reactor table reactor device ---@param rps rps RPS reference ---@param conn_watchdog watchdog watchdog reference function plc.comms(id, version, modem, local_port, server_port, range, reactor, rps, conn_watchdog) local self = { seq_num = 0, r_seq_num = nil, scrammed = false, linked = false, last_est_ack = ESTABLISH_ACK.ALLOW, resend_build = false, auto_ack_token = 0, status_cache = nil, max_burn_rate = nil } comms.set_trusted_range(range) -- PRIVATE FUNCTIONS -- -- configure modem channels local function _conf_channels() modem.closeAll() modem.open(local_port) end _conf_channels() -- send an RPLC packet ---@param msg_type RPLC_TYPE ---@param msg table local function _send(msg_type, msg) local s_pkt = comms.scada_packet() local r_pkt = comms.rplc_packet() r_pkt.make(id, msg_type, msg) s_pkt.make(self.seq_num, PROTOCOL.RPLC, r_pkt.raw_sendable()) modem.transmit(server_port, local_port, s_pkt.raw_sendable()) self.seq_num = self.seq_num + 1 end -- send a SCADA management packet ---@param msg_type SCADA_MGMT_TYPE ---@param msg table local function _send_mgmt(msg_type, msg) local s_pkt = comms.scada_packet() local m_pkt = comms.mgmt_packet() m_pkt.make(msg_type, msg) s_pkt.make(self.seq_num, PROTOCOL.SCADA_MGMT, m_pkt.raw_sendable()) modem.transmit(server_port, local_port, s_pkt.raw_sendable()) self.seq_num = self.seq_num + 1 end -- variable reactor status information, excluding heating rate ---@return table data_table, boolean faulted local function _reactor_status() local fuel = nil local waste = nil local coolant = nil local hcoolant = nil local data_table = { false, -- getStatus 0, -- getBurnRate 0, -- getActualBurnRate 0, -- getTemperature 0, -- getDamagePercent 0, -- getBoilEfficiency 0, -- getEnvironmentalLoss 0, -- fuel_amnt 0, -- getFuelFilledPercentage 0, -- waste_amnt 0, -- getWasteFilledPercentage "", -- coolant_name 0, -- coolant_amnt 0, -- getCoolantFilledPercentage "", -- hcoolant_name 0, -- hcoolant_amnt 0 -- getHeatedCoolantFilledPercentage } local tasks = { function () data_table[1] = reactor.getStatus() end, function () data_table[2] = reactor.getBurnRate() end, function () data_table[3] = reactor.getActualBurnRate() end, function () data_table[4] = reactor.getTemperature() end, function () data_table[5] = reactor.getDamagePercent() end, function () data_table[6] = reactor.getBoilEfficiency() end, function () data_table[7] = reactor.getEnvironmentalLoss() end, function () fuel = reactor.getFuel() end, function () data_table[9] = reactor.getFuelFilledPercentage() end, function () waste = reactor.getWaste() end, function () data_table[11] = reactor.getWasteFilledPercentage() end, function () coolant = reactor.getCoolant() end, function () data_table[14] = reactor.getCoolantFilledPercentage() end, function () hcoolant = reactor.getHeatedCoolant() end, function () data_table[17] = reactor.getHeatedCoolantFilledPercentage() end } parallel.waitForAll(table.unpack(tasks)) if fuel ~= nil then data_table[8] = fuel.amount end if waste ~= nil then data_table[10] = waste.amount end if coolant ~= nil then data_table[12] = coolant.name data_table[13] = coolant.amount end if hcoolant ~= nil then data_table[15] = hcoolant.name data_table[16] = hcoolant.amount end return data_table, reactor.__p_is_faulted() end -- update the status cache if changed ---@return boolean changed local function _update_status_cache() local status, faulted = _reactor_status() local changed = false if self.status_cache ~= nil then if not faulted then for i = 1, #status do if status[i] ~= self.status_cache[i] then changed = true break end end end else changed = true end if changed and not faulted then self.status_cache = status end return changed end -- keep alive ack ---@param srv_time integer local function _send_keep_alive_ack(srv_time) _send_mgmt(SCADA_MGMT_TYPE.KEEP_ALIVE, { srv_time, util.time() }) end -- general ack ---@param msg_type RPLC_TYPE ---@param status boolean|integer local function _send_ack(msg_type, status) _send(msg_type, { status }) end -- send structure properties (these should not change, server will cache these) local function _send_struct() local min_pos = { x = 0, y = 0, z = 0 } local max_pos = { x = 0, y = 0, z = 0 } local mek_data = { false, 0, 0, 0, min_pos, max_pos, 0, 0, 0, 0, 0, 0, 0, 0 } local tasks = { function () mek_data[1] = reactor.getLength() end, function () mek_data[2] = reactor.getWidth() end, function () mek_data[3] = reactor.getHeight() end, function () mek_data[4] = reactor.getMinPos() end, function () mek_data[5] = reactor.getMaxPos() end, function () mek_data[6] = reactor.getHeatCapacity() end, function () mek_data[7] = reactor.getFuelAssemblies() end, function () mek_data[8] = reactor.getFuelSurfaceArea() end, function () mek_data[9] = reactor.getFuelCapacity() end, function () mek_data[10] = reactor.getWasteCapacity() end, function () mek_data[11] = reactor.getCoolantCapacity() end, function () mek_data[12] = reactor.getHeatedCoolantCapacity() end, function () mek_data[13] = reactor.getMaxBurnRate() end } parallel.waitForAll(table.unpack(tasks)) if not reactor.__p_is_faulted() then _send(RPLC_TYPE.MEK_STRUCT, mek_data) self.resend_build = false end end -- PUBLIC FUNCTIONS -- ---@class plc_comms local public = {} -- reconnect a newly connected modem ---@param new_modem table function public.reconnect_modem(new_modem) modem = new_modem _conf_channels() end -- reconnect a newly connected reactor ---@param new_reactor table function public.reconnect_reactor(new_reactor) reactor = new_reactor self.status_cache = nil self.resend_build = true self.max_burn_rate = nil end -- unlink from the server function public.unlink() self.linked = false self.r_seq_num = nil self.status_cache = nil end -- close the connection to the server function public.close() conn_watchdog.cancel() public.unlink() _send_mgmt(SCADA_MGMT_TYPE.CLOSE, {}) end -- attempt to establish link with supervisor function public.send_link_req() _send_mgmt(SCADA_MGMT_TYPE.ESTABLISH, { comms.version, version, DEVICE_TYPE.PLC, id }) end -- send live status information ---@param no_reactor boolean PLC lost reactor connection ---@param formed boolean reactor formed (from PLC state) function public.send_status(no_reactor, formed) if self.linked then local mek_data = nil ---@type table local heating_rate = 0.0 ---@type number if (not no_reactor) and rps.is_formed() then if _update_status_cache() then mek_data = self.status_cache end heating_rate = reactor.getHeatingRate() end local sys_status = { util.time(), -- timestamp (not self.scrammed), -- requested control state no_reactor, -- no reactor peripheral connected formed, -- reactor formed self.auto_ack_token, -- token to indicate auto command has been received before this status update heating_rate, -- heating rate mek_data -- mekanism status data } _send(RPLC_TYPE.STATUS, sys_status) if self.resend_build then _send_struct() end end end -- send reactor protection system status function public.send_rps_status() if self.linked then _send(RPLC_TYPE.RPS_STATUS, { rps.is_tripped(), rps.get_trip_cause(), table.unpack(rps.status()) }) end end -- send reactor protection system alarm ---@param cause rps_trip_cause reactor protection system status function public.send_rps_alarm(cause) if self.linked then _send(RPLC_TYPE.RPS_ALARM, { cause, table.unpack(rps.status()) }) end end -- parse an RPLC packet ---@nodiscard ---@param side string ---@param sender integer ---@param reply_to integer ---@param message any ---@param distance integer ---@return rplc_frame|mgmt_frame|nil packet function public.parse_packet(side, sender, reply_to, message, distance) local pkt = nil local s_pkt = comms.scada_packet() -- parse packet as generic SCADA packet s_pkt.receive(side, sender, reply_to, message, distance) if s_pkt.is_valid() then -- get as RPLC packet if s_pkt.protocol() == PROTOCOL.RPLC then local rplc_pkt = comms.rplc_packet() if rplc_pkt.decode(s_pkt) then pkt = rplc_pkt.get() end -- get as SCADA management packet elseif s_pkt.protocol() == PROTOCOL.SCADA_MGMT then local mgmt_pkt = comms.mgmt_packet() if mgmt_pkt.decode(s_pkt) then pkt = mgmt_pkt.get() end else log.debug("illegal packet type " .. s_pkt.protocol(), true) end end return pkt end -- handle an RPLC packet ---@param packet rplc_frame|mgmt_frame packet frame ---@param plc_state plc_state PLC state ---@param setpoints setpoints setpoint control table function public.handle_packet(packet, plc_state, setpoints) -- print a log message to the terminal as long as the UI isn't running local function println_ts(message) if not plc_state.fp_ok then util.println_ts(message) end end -- handle packets now that we have prints setup if packet.scada_frame.local_port() == local_port then -- check sequence number if self.r_seq_num == nil then self.r_seq_num = packet.scada_frame.seq_num() elseif self.linked and self.r_seq_num >= packet.scada_frame.seq_num() then log.warning("sequence out-of-order: last = " .. self.r_seq_num .. ", new = " .. packet.scada_frame.seq_num()) return else self.r_seq_num = packet.scada_frame.seq_num() end -- feed the watchdog first so it doesn't uhh...eat our packets :) conn_watchdog.feed() local protocol = packet.scada_frame.protocol() -- handle packet if protocol == PROTOCOL.RPLC then ---@cast packet rplc_frame if self.linked then if packet.type == RPLC_TYPE.STATUS then -- request of full status, clear cache first self.status_cache = nil public.send_status(plc_state.no_reactor, plc_state.reactor_formed) log.debug("sent out status cache again, did supervisor miss it?") elseif packet.type == RPLC_TYPE.MEK_STRUCT then -- request for physical structure _send_struct() log.debug("sent out structure again, did supervisor miss it?") elseif packet.type == RPLC_TYPE.MEK_BURN_RATE then -- set the burn rate if (packet.length == 2) and (type(packet.data[1]) == "number") then local success = false local burn_rate = math.floor(packet.data[1] * 10) / 10 local ramp = packet.data[2] -- if no known max burn rate, check again if self.max_burn_rate == nil then self.max_burn_rate = reactor.getMaxBurnRate() end -- if we know our max burn rate, update current burn rate setpoint if in range if self.max_burn_rate ~= ppm.ACCESS_FAULT then if burn_rate > 0 and burn_rate <= self.max_burn_rate then if ramp then setpoints.burn_rate_en = true setpoints.burn_rate = burn_rate success = true else reactor.setBurnRate(burn_rate) success = not reactor.__p_is_faulted() end else log.debug(burn_rate .. " rate outside of 0 < x <= " .. self.max_burn_rate) end end _send_ack(packet.type, success) else log.debug("RPLC set burn rate packet length mismatch or non-numeric burn rate") end elseif packet.type == RPLC_TYPE.RPS_ENABLE then -- enable the reactor self.scrammed = false _send_ack(packet.type, rps.activate()) elseif packet.type == RPLC_TYPE.RPS_SCRAM then -- disable the reactor per manual request self.scrammed = true rps.trip_manual() _send_ack(packet.type, true) elseif packet.type == RPLC_TYPE.RPS_ASCRAM then -- disable the reactor per automatic request self.scrammed = true rps.trip_auto() _send_ack(packet.type, true) elseif packet.type == RPLC_TYPE.RPS_RESET then -- reset the RPS status rps.reset() _send_ack(packet.type, true) elseif packet.type == RPLC_TYPE.RPS_AUTO_RESET then -- reset automatic SCRAM and timeout trips rps.auto_reset() _send_ack(packet.type, true) elseif packet.type == RPLC_TYPE.AUTO_BURN_RATE then -- automatic control requested a new burn rate if (packet.length == 3) and (type(packet.data[1]) == "number") and (type(packet.data[3]) == "number") then local ack = AUTO_ACK.FAIL local burn_rate = math.floor(packet.data[1] * 100) / 100 local ramp = packet.data[2] self.auto_ack_token = packet.data[3] -- if no known max burn rate, check again if self.max_burn_rate == nil then self.max_burn_rate = reactor.getMaxBurnRate() end -- if we know our max burn rate, update current burn rate setpoint if in range if self.max_burn_rate ~= ppm.ACCESS_FAULT then if burn_rate < 0.01 then if rps.is_active() then -- auto scram to disable log.debug("AUTO: stopping the reactor to meet 0.0 burn rate") if rps.scram() then ack = AUTO_ACK.ZERO_DIS_OK else log.warning("AUTO: automatic reactor stop failed") end else ack = AUTO_ACK.ZERO_DIS_OK end elseif burn_rate <= self.max_burn_rate then if not rps.is_active() then -- activate the reactor log.debug("AUTO: activating the reactor") reactor.setBurnRate(0.01) if reactor.__p_is_faulted() then log.warning("AUTO: failed to reset burn rate for auto activation") else if not rps.auto_activate() then log.warning("AUTO: automatic reactor activation failed") end end end -- if active, set/ramp burn rate if rps.is_active() then if ramp then log.debug(util.c("AUTO: setting burn rate ramp to ", burn_rate)) setpoints.burn_rate_en = true setpoints.burn_rate = burn_rate ack = AUTO_ACK.RAMP_SET_OK else log.debug(util.c("AUTO: setting burn rate directly to ", burn_rate)) reactor.setBurnRate(burn_rate) ack = util.trinary(reactor.__p_is_faulted(), AUTO_ACK.FAIL, AUTO_ACK.DIRECT_SET_OK) end end else log.debug(util.c(burn_rate, " rate outside of 0 < x <= ", self.max_burn_rate)) end end _send_ack(packet.type, ack) else log.debug("RPLC set automatic burn rate packet length mismatch or non-numeric burn rate") end else log.debug("received unknown RPLC packet type " .. packet.type) end else log.debug("discarding RPLC packet before linked") end elseif protocol == PROTOCOL.SCADA_MGMT then ---@cast packet mgmt_frame if self.linked then if packet.type == SCADA_MGMT_TYPE.ESTABLISH then -- link request confirmation if packet.length == 1 then log.debug("received unsolicited establish response") local est_ack = packet.data[1] if est_ack == ESTABLISH_ACK.ALLOW then self.status_cache = nil _send_struct() public.send_status(plc_state.no_reactor, plc_state.reactor_formed) log.debug("re-sent initial status data") elseif est_ack == ESTABLISH_ACK.DENY then println_ts("received unsolicited link denial, unlinking") log.warning("unsolicited establish request denied") elseif est_ack == ESTABLISH_ACK.COLLISION then println_ts("received unsolicited link collision, unlinking") log.warning("unsolicited establish request collision") elseif est_ack == ESTABLISH_ACK.BAD_VERSION then println_ts("received unsolicited link version mismatch, unlinking") log.warning("unsolicited establish request version mismatch") else println_ts("invalid unsolicited link response") log.debug("unsolicited unknown establish request response") end self.linked = est_ack == ESTABLISH_ACK.ALLOW -- clear this since this is for something that was unsolicited self.last_est_ack = ESTABLISH_ACK.ALLOW -- report link state databus.tx_link_state(est_ack + 1) else log.debug("SCADA_MGMT establish packet length mismatch") end elseif packet.type == SCADA_MGMT_TYPE.KEEP_ALIVE then -- keep alive request received, echo back if packet.length == 1 and type(packet.data[1]) == "number" then local timestamp = packet.data[1] local trip_time = util.time() - timestamp if trip_time > 750 then log.warning("PLC KEEP_ALIVE trip time > 750ms (" .. trip_time .. "ms)") end -- log.debug("RPLC RTT = " .. trip_time .. "ms") _send_keep_alive_ack(timestamp) else log.debug("SCADA_MGMT keep alive packet length/type mismatch") end elseif packet.type == SCADA_MGMT_TYPE.CLOSE then -- handle session close conn_watchdog.cancel() public.unlink() println_ts("server connection closed by remote host") log.warning("server connection closed by remote host") else log.debug("received unsupported SCADA_MGMT packet type " .. packet.type) end elseif packet.type == SCADA_MGMT_TYPE.ESTABLISH then -- link request confirmation if packet.length == 1 then local est_ack = packet.data[1] if est_ack == ESTABLISH_ACK.ALLOW then println_ts("linked!") log.info("supervisor establish request approved, PLC is linked") -- reset remote sequence number and cache self.r_seq_num = nil self.status_cache = nil if plc_state.reactor_formed then _send_struct() end public.send_status(plc_state.no_reactor, plc_state.reactor_formed) log.debug("sent initial status data") elseif self.last_est_ack ~= est_ack then if est_ack == ESTABLISH_ACK.DENY then println_ts("link request denied, retrying...") log.info("supervisor establish request denied, retrying") elseif est_ack == ESTABLISH_ACK.COLLISION then println_ts("reactor PLC ID collision (check config), retrying...") log.warning("establish request collision, retrying") elseif est_ack == ESTABLISH_ACK.BAD_VERSION then println_ts("supervisor version mismatch (try updating), retrying...") log.warning("establish request version mismatch, retrying") else println_ts("invalid link response, bad channel? retrying...") log.error("unknown establish request response, retrying") end end self.linked = est_ack == ESTABLISH_ACK.ALLOW self.last_est_ack = est_ack -- report link state databus.tx_link_state(est_ack + 1) else log.debug("SCADA_MGMT establish packet length mismatch") end else log.debug("discarding non-link SCADA_MGMT packet before linked") end else -- should be unreachable assuming packet is from parse_packet() log.error("illegal packet type " .. protocol, true) end end end ---@nodiscard function public.is_scrammed() return self.scrammed end ---@nodiscard function public.is_linked() return self.linked end return public end return plc