ntopng/scripts/lua/modules/timeseries/drivers/influxdb.lua

--
-- (C) 2018 - ntop.org
--

local driver = {}

-- NOTE: this script is required by second.lua, keep the imports minimal!
local ts_common = require("ts_common")
local json = require("dkjson")
local os_utils = require("os_utils")
require("ntop_utils")

--
-- Sample query:
--    select * from "iface:ndpi" where ifid='0' and protocol='SSL'
--
-- See also callback_utils.uploadTSdata
--

-- Export problems occur if the export queue length is >= INFLUX_MAX_EXPORT_QUEUE_LEN_HIGH.
-- Then, the problems are considered fixed once the queue length goes <= INFLUX_MAX_EXPORT_QUEUE_LEN_LOW.
-- NOTE: these values are multiplied by the number of interfaces
-- NOTE: a single file can be as big as 4MB.
local INFLUX_MAX_EXPORT_QUEUE_LEN_LOW = 10
local INFLUX_MAX_EXPORT_QUEUE_LEN_HIGH = 20
local INFLUX_MAX_EXPORT_QUEUE_TRIM_LEN = 30 -- This edge should never be crossed. If it does, queue is manually trimmed

local INFLUX_EXPORT_QUEUE = "ntopng.influx_file_queue"
local MIN_INFLUXDB_SUPPORTED_VERSION = "1.5.1"
local FIRST_AGGREGATION_TIME_KEY = "ntopng.prefs.influxdb.first_aggregation_time"

-- hourly continuous queries are disabled as they create a lot of pressure
-- on the InfluxDB database.
local HOURLY_CQ_ENABLED = false
local HOURLY_CQ_DISABLED_KEY = "ntopng.prefs.influxdb.1h_cq_disabled"

-- ##############################################

local INFLUX_KEY_PREFIX = "ntopng.cache.influxdb."

-- Keep total counters for dropped points
local INFLUX_KEY_DROPPED_POINTS = INFLUX_KEY_PREFIX.."num_dropped_points"
local INFLUX_KEY_EXPORTED_POINTS = INFLUX_KEY_PREFIX.."num_exported_points"
local INFLUX_KEY_EXPORTS = INFLUX_KEY_PREFIX.."num_exports"
local INFLUX_KEY_FAILED_EXPORTS = INFLUX_KEY_PREFIX.."num_failed_exports"

-- Use this flag as TTL-based redis keys to check wether the health
-- of influxdb is OK.
local INFLUX_FLAGS_TIMEOUT = 60 -- keep the issue for 60 seconds
local INFLUX_FLAG_DROPPING_POINTS = INFLUX_KEY_PREFIX.."flag_dropping_points"
local INFLUX_FLAG_FAILING_EXPORTS = INFLUX_KEY_PREFIX.."flag_failing_exports"

local INFLUX_FLAG_IS_CURRENTLY_DROPPING = INFLUX_KEY_PREFIX.."flag_currently_dropping"
local INFLUX_QUEUE_FULL_FLAG = INFLUX_KEY_PREFIX.."export_queue_full"

-- ##############################################

local function isExportQueueFull()
  return(ntop.getCache(INFLUX_QUEUE_FULL_FLAG) == "1")
end

local function setExportQueueFull(is_full)
   if(is_full) then
      ntop.setCache(INFLUX_QUEUE_FULL_FLAG, "1")
   else
      ntop.delCache(INFLUX_QUEUE_FULL_FLAG)
   end
end

-- ##############################################

function driver:new(options)
  local obj = {
    url = options.url,
    db = options.db,
    username = options.username or "",
    password = options.password or "",
    has_full_export_queue = isExportQueueFull(),
    cur_dropped_points = 0,
  }

  setmetatable(obj, self)
  self.__index = self

  return obj
end

-- ##############################################

local function getInfluxDBQueryTimeout()
  return tonumber(ntop.getPref("ntopng.prefs.influx_query_timeout")) or 10
end

-- ##############################################

local function inc_dropped_points(num_points)
   ntop.incrCache(INFLUX_KEY_DROPPED_POINTS, num_points)

   -- Status flag
   ntop.setCache(INFLUX_FLAG_DROPPING_POINTS, "true", INFLUX_FLAGS_TIMEOUT)

   -- Internal flag
   ntop.setCache(INFLUX_FLAG_IS_CURRENTLY_DROPPING, "true")
end

-- ##############################################

local function inc_exported_points(num_points)
  ntop.incrCache(INFLUX_KEY_EXPORTED_POINTS, num_points)
end

-- ##############################################

local function inc_exports()
   ntop.incrCache(INFLUX_KEY_EXPORTS)
end

-- ##############################################

local function inc_failed_exports()
   ntop.setCache(INFLUX_FLAG_FAILING_EXPORTS, "true", INFLUX_FLAGS_TIMEOUT)
   ntop.incrCache(INFLUX_KEY_FAILED_EXPORTS, 1)
end

-- ##############################################

local function del_all_vals()
   ntop.delCache(INFLUX_KEY_DROPPED_POINTS)
   ntop.delCache(INFLUX_KEY_EXPORTED_POINTS)
   ntop.delCache(INFLUX_KEY_EXPORTS)
   ntop.delCache(INFLUX_KEY_FAILED_EXPORTS)
   ntop.delCache(INFLUX_FLAG_DROPPING_POINTS)
   ntop.delCache(INFLUX_FLAG_FAILING_EXPORTS)
   ntop.delCache(INFLUX_FLAG_IS_CURRENTLY_DROPPING)
end

-- ##############################################

function driver:get_dropped_points()
   return tonumber(ntop.getCache(INFLUX_KEY_DROPPED_POINTS)) or 0
end

-- ##############################################

function driver:get_exported_points()
   return tonumber(ntop.getCache(INFLUX_KEY_EXPORTED_POINTS)) or 0
end

-- ##############################################

function driver:get_exports()
   return tonumber(ntop.getCache(INFLUX_KEY_EXPORTS)) or 0
end

-- ##############################################

local function is_dropping_points()
   return ntop.getCache(INFLUX_FLAG_DROPPING_POINTS) == "true"
end

-- ##############################################

local function is_failing_exports()
   return ntop.getCache(INFLUX_FLAG_FAILING_EXPORTS) == "true"
end

-- ##############################################

-- This function is called when the object is destroyed (garbage collected)
function driver:__gc()
  if(self.cur_dropped_points > 0) then
    inc_dropped_points(self.cur_dropped_points)
  end
end

-- ##############################################

function driver:append(schema, timestamp, tags, metrics)
  if(self.has_full_export_queue) then
    -- Temporary buffer dropped points into a local counter.
    -- They will be exported to redis once the driver is dismissed in driver:__gc()
    self.cur_dropped_points = self.cur_dropped_points + 1
    return(false)
  end

  local rv = interface.appendInfluxDB(schema.name, timestamp, tags, metrics)

  if not rv then
     -- Unable to append the point for export
     self.cur_dropped_points = self.cur_dropped_points + 1
  end

  return rv
end

-- ##############################################

local function getResponseError(res)
  if res.CONTENT and res.CONTENT_TYPE == "application/json" then
    local jres = json.decode(res.CONTENT)

    if jres then
      if jres.error then
        if res.RESPONSE_CODE then
          return "[".. res.RESPONSE_CODE .. "] " ..jres.error
        else
          return jres.error
        end
      elseif jres.results then
        for _, single_res in pairs(jres.results) do
          if single_res.error then
            return single_res.error, single_res.statement_id
          end
        end
      end
    end
  end

  return res.RESPONSE_CODE or "unknown error"
end

-- ##############################################

local function getDatabaseRetentionDays()
   local data_retention_utils = require "data_retention_utils"
   return data_retention_utils.getDataRetentionDays()
end

local function get1dDatabaseRetentionDays()
  return getDatabaseRetentionDays()
end

local function get1hDatabaseRetentionDays()
  return getDatabaseRetentionDays()
end

-- ##############################################

local function isRollupEnabled()
  return(ntop.getPref("ntopng.prefs.disable_influxdb_rollup") ~= "1")
end

 --##############################################

-- Determines the most appropriate retention policy
local function getSchemaRetentionPolicy(schema, tstart, tend, options)
  if schema.options.influx_internal_query then
    return "raw"
  end

  options = options or {}
  local first_aggr_time = tonumber(ntop.getPref(FIRST_AGGREGATION_TIME_KEY))

  if((not first_aggr_time) or (not isRollupEnabled())) then
    return "raw"
  end

  local rp_1d_duration_sec = get1dDatabaseRetentionDays() * 86400
  local rp_1h_duration_sec = get1hDatabaseRetentionDays() * 86400

  -- RP selection logic
  local oldest_1d_data = os.time() - rp_1d_duration_sec
  local oldest_1h_data = os.time() - rp_1h_duration_sec
  local oldest_raw_data = getDatabaseRetentionDays() * 86400
  local max_raw_interval = 12 * 3600 -- after 12 hours begin to use the aggregated data
  local max_1h_interval

  if(HOURLY_CQ_ENABLED) then
    max_1h_interval = 15 * 86400 -- after 15 days use the 1d aggregated data
  else
    max_1h_interval = 5 * 86400 -- after 5 days use the 1d aggregated data
  end

  if options.target_aggregation then
    if((options.target_aggregation == "1h") and (tstart < oldest_1h_data)) or
      ((options.target_aggregation == "raw") and (tstart < oldest_raw_data)) then
      -- cannot satisfy the target_aggregation
    else
      -- user override
      return options.target_aggregation
    end
  end

  -- Fix the limits to ensure that the first_aggr_time is not crossed
  -- Crossing it is still acceptable for the options.target_aggregation check above
  oldest_1d_data = math.max(oldest_1d_data, first_aggr_time)
  oldest_1h_data = math.max(oldest_1h_data, first_aggr_time)

  if tstart < oldest_1d_data then
    return "raw"
  elseif tstart < oldest_1h_data then
    return "1d"
  elseif(HOURLY_CQ_ENABLED and (tstart < oldest_raw_data)) then
    return "1h"
  end

  local interval = tend - tstart

  if interval >= max_1h_interval then
    return "1d"
  elseif(HOURLY_CQ_ENABLED and (interval >= max_raw_interval)) then
    return "1h"
  end

  return "raw"
end

-- ##############################################

-- This is necessary to avoid fetching a point which is not already
-- aggregated by the retention policy
local function fixTendForRetention(tend, rp)
  local now = os.time()

  if(rp == "1h") then
    local current_hour = now - (now % 3600)
    return(math.min(tend, current_hour - 3600))
  elseif(rp == "1d") then
    local current_day = os.date("*t", now)
    current_day.min = 0
    current_day.sec = 0
    current_day.hour = 0
    current_day = os.time(current_day)
    return(math.min(tend, current_day - 86400))
  end

  return(tend)
end

-- ##############################################

-- returns schema_name, step
local function retentionPolicyToSchema(schema, rp, db)
  if((rp == "raw") or (rp == "autogen")) then
    rp = nil
  end

  if not isEmptyString(rp) then
    -- counters become derivatives
    local metric_type = ternary(schema.options.metrics_type == ts_common.metrics.counter, ts_common.metrics.derivative, schema.options.metrics_type)
    return string.format('"%s"."%s"."%s"', db, rp, schema.name), ternary(rp == "1d", 86400, 3600), metric_type
  end

  -- raw
  return string.format('"%s"', schema.name), schema.options.step, schema.options.metrics_type
end

-- ##############################################

local function getQuerySchema(schema, tstart, tend, db, options)
  return retentionPolicyToSchema(schema, getSchemaRetentionPolicy(schema, tstart, tend, options), db)
end

-- ##############################################

local function influx_query_multi(base_url, query, username, password, options)
  options = options or {}
  local full_url = base_url .."&q=" .. urlencode(query)
  local tstart = os.time()
  local timeout = ternary(options.no_timeout, 99999999999, getInfluxDBQueryTimeout())
  local res = ntop.httpGet(full_url, username, password, timeout, true)
  local tdiff = os.time() - tstart
  local debug_influxdb_queries = (ntop.getPref("ntopng.prefs.influxdb.print_queries") == "1")

  if debug_influxdb_queries then
     local _, num_queries = string.gsub(query, ";", "")
     traceError(TRACE_NORMAL, TRACE_CONSOLE, string.format("influx_query[#%u][%ds]: %s", num_queries+1, tdiff, query))
  end

  if tdiff >= timeout then
    -- Show the error
    ntop.setCache("ntopng.cache.influxdb.last_error", i18n("graphs.influxdb_not_responding", {
      url = ntop.getHttpPrefix() .. "/lua/admin/prefs.lua?tab=on_disk_ts#influx_query_timeout",
      flask_icon = '<i class="fas fa-flask"></i>',
    }))
  end

  -- Log the query
  local msg = os.date("%d/%b/%Y %H:%M:%S ") .. query
  ntop.lpushCache("ntopng.trace.influxdb_queries", msg, 100 --[[ max queue elements ]])

  if not res then
    traceError(TRACE_ERROR, TRACE_CONSOLE, "Invalid response for query: " .. full_url)
    return nil
  end

  if((res.RESPONSE_CODE ~= 200) and (not res.IS_PARTIAL)) then
    traceError(TRACE_ERROR, TRACE_CONSOLE, "Bad response code[" .. res.RESPONSE_CODE .. "]: " .. (res.CONTENT or ""))
    return nil
  end

  if((res.CONTENT == nil) or (res.IS_PARTIAL)) then
    ts_common.setLastError(ts_common.ERR_OPERATION_TOO_SLOW, i18n("graphs.query_too_slow"))
    return nil
  end

  local jres = json.decode(res.CONTENT)

  if (not jres) or (not jres.results) or (not #jres.results) then
    traceError(TRACE_ERROR, TRACE_CONSOLE, "Invalid JSON reply[" .. res.CONTENT_LEN .. " bytes]: " .. string.sub(res.CONTENT, 1, 50))
    return nil
  end

  for _, result in ipairs(jres.results) do
    if result["error"] then
      traceError(TRACE_ERROR, TRACE_CONSOLE, result["error"])
      return nil
    end
  end

  return jres
end

-- ##############################################

local function influx_query(base_url, query, username, password, options)
  local jres = influx_query_multi(base_url, query, username, password, options)

  if jres and jres.results then
    if not jres.results[1].series then
      -- no results found
      return {}
    end

    return jres.results[1]
  end

  return nil
end

-- ##############################################

local function multiQueryPost(queries, url, username, password)
  local query_str = table.concat(queries, ";")
  local res = ntop.httpPost(url .. "/query", "q=" .. urlencode(query_str), username, password, getInfluxDBQueryTimeout(), true)

  if not res then
    local err = "Invalid response for query: " .. query_str
    traceError(TRACE_ERROR, TRACE_CONSOLE, err)
    return false, err
  end

  if res.RESPONSE_CODE ~= 200 then
    local err = "Bad response code[" .. res.RESPONSE_CODE .. "]: " .. getResponseError(res)
    traceError(TRACE_ERROR, TRACE_CONSOLE, err)
    --tprint(query_str)
    return false, err
  end

  local err, statement_id = getResponseError(res)
  if err ~= 200 then
    local err = "Unexpected query error: " .. err
    if statement_id ~= nil then
      err = err .. string.format(", in query #%d: %s", statement_id, queries[statement_id+1] or "nil")
    end
    traceError(TRACE_ERROR, TRACE_CONSOLE, err)
    return false, err
  end

  return true
end

-- ##############################################

local function influx2Series(schema, tstart, tend, tags, options, data, time_step, last_ts)
  local series = {}
  local max_vals = {}
  last_ts = last_ts or os.time()

  -- Create the columns
  for i=2, #data.columns do
    series[i-1] = {label=data.columns[i], data={}}
    max_vals[i-1] = ts_common.getMaxPointValue(schema, series[i-1].label, tags)
  end

  -- The first time available in the returned data
  local first_t = data.values[1][1]
  -- Align tstart to the first timestamp
  tstart = tstart + (first_t - tstart) % time_step

  -- next_t holds the expected timestamp of the next point to process
  local next_t = tstart
  -- the next index to use for insertion in the result table
  local series_idx = 1
  --tprint(time_step .. ") " .. tstart .. " vs " .. first_t .. " - " .. next_t)

  -- Convert the data
  for idx, values in ipairs(data.values) do
    local cur_t = data.values[idx][1]

    if #values < 2 or cur_t < tstart then
      -- skip empty points (which are out of query bounds)
      goto continue
    end

    if(cur_t > last_ts) then
      -- skip values exceeding the last point
      break
    end

    -- Fill the missing points
    while((cur_t - next_t) >= time_step) do
      for _, serie in pairs(series) do
        serie.data[series_idx] = options.fill_value
      end

      --tprint("FILL [" .. series_idx .."] " .. cur_t .. " vs " .. next_t)
      series_idx = series_idx + 1
      next_t = next_t + time_step
    end

    for i=2, #values do
      local val = ts_common.normalizeVal(values[i], max_vals[i-1], options)
      series[i-1].data[series_idx] = val
    end

    --traceError(TRACE_NORMAL, TRACE_CONSOLE, string.format("@ %u = %.2f", cur_t, values[2]))
    if(false) then -- consinstency check
      local expected_t = next_t
      local actual_t = values[1]

      if math.abs(expected_t - actual_t) >= time_step then
        traceError(TRACE_WARNING, TRACE_CONSOLE,
          string.format("Bad point timestamp: expected %u, found %u [value = %.2f]", expected_t, actual_t, values[2]))
      end
    end

    series_idx = series_idx + 1
    next_t = next_t + time_step

    ::continue::
  end

   -- Fill the missing points at the end
  while((tend - next_t) >= 0) do
    -- NOTE: fill_series is required for composed charts. In such case,
    -- not filling the serie would result into an incorrect upsampling
    if((not options.fill_series) and (next_t > last_ts)) then
      -- skip values exceeding the last point
      break
    end

    for _, serie in pairs(series) do
      serie.data[series_idx] = options.fill_value
    end

    --tprint("FILL [" .. series_idx .."] " .. tend .. " vs " .. next_t)
    series_idx = series_idx + 1
    next_t = next_t + time_step
  end

  local count = series_idx - 1

  return series, count, tstart
end

-- Test only
driver._influx2Series = influx2Series

 --##############################################

local function where_tags(tags)
  if not table.empty(tags) then
    return ' WHERE ' .. table.tconcat(tags, "=", " AND ", nil, "'") .. " AND"
  else
    return " WHERE"
  end
end

 --##############################################

local function getTotalSerieQuery(schema, query_schema, raw_step, tstart, tend, tags, time_step, data_type, label)
  label = label or "total_serie"

  --[[
  SELECT NON_NEGATIVE_DERIVATIVE(total_serie) AS total_serie FROM               // derivate the serie, if necessary
    (SELECT MEAN("total_serie") AS "total_serie" FROM                           // sample the total serie points, if necessary
      (SELECT SUM("value") AS "total_serie" FROM                                // sum all the series together
        (SELECT (bytes_sent + bytes_rcvd) AS "value" FROM "host:ndpi"           // possibly sum multiple metrics within same serie
          WHERE host='192.168.43.18' AND ifid='2'
          AND time >= 1531916170000000000 AND time <= 1532002570000000000)
        GROUP BY time(300s))
      GROUP BY time(600s))
  ]]
  local is_single_serie = schema:allTagsDefined(tags)
  local simplified_query = 'SELECT (' .. table.concat(schema._metrics, " + ") ..') AS "'.. label ..'" FROM '.. query_schema .. where_tags(tags) .. ' time >= ' .. tstart .. '000000000 AND time <= ' .. tend .. '000000000'

  local query

  if is_single_serie then
    -- optimized version
    query = simplified_query
  else
    query = 'SELECT SUM("'.. label ..'") AS "'.. label ..'" FROM ('.. simplified_query ..')'..
      ' GROUP BY time('.. raw_step ..'s)'
  end

  if time_step and (raw_step ~= time_step) then
    -- sample the points
    query = 'SELECT MEAN("'.. label ..'") AS "'.. label ..'" FROM ('.. query ..') GROUP BY time('.. time_step ..'s)'
  end

  if data_type == ts_common.metrics.counter then
    local optimized = false

    if (query == simplified_query) and (#schema._metrics == 1) then
      -- Remove nested query if possible
      local parts = split(query, " FROM ")

      if #parts == 2 then
        query = "SELECT NON_NEGATIVE_DERIVATIVE(".. schema._metrics[1] ..") AS ".. label .." FROM " .. parts[2]
      end
    end

    if not optimized then
      query = "SELECT NON_NEGATIVE_DERIVATIVE(".. label ..") AS ".. label .." FROM (" .. query .. ")"
    end
  end

  return query
end

-- ##############################################

local function getLastTsQuery(schema, query_schema, tags)
  return string.format('SELECT LAST('.. schema._metrics[1] ..') FROM %s WHERE 1=1 AND %s', query_schema, table.tconcat(tags, "=", " AND ", nil, "'"))
end

-- ##############################################

function driver:_makeTotalSerie(schema, query_schema, raw_step, tstart, tend, tags, options, url, time_step, label, unaligned_offset, data_type)
  local query = getTotalSerieQuery(schema, query_schema, raw_step, tstart, tend + unaligned_offset, tags, time_step, data_type, label)
  local last_ts_query = getLastTsQuery(schema, query_schema, tags)
  local jres = influx_query_multi(url .. "/query?db=".. self.db .."&epoch=s", string.format("%s;%s", query, last_ts_query), self.username, self.password, options)
  local last_ts = os.time()
  local data = {}

  if(jres and jres.results and (#jres.results == 2)) then
    if jres.results[1].series then
      data = jres.results[1]
    end

    if jres.results[2].series and jres.results[2].series[1].values then
      last_ts = jres.results[2].series[1].values[1][1]
    end
  end

  if table.empty(data) then
    local rv = {}
    local i = 1

    for t=tstart + time_step, tend, time_step do
      rv[i] = 0
      i = i + 1
    end

    return rv
  end

  data = data.series[1]

  local series, count, tstart = influx2Series(schema, tstart + time_step, tend, tags, options, data, time_step, last_ts)
  return series[1].data
end

-- ##############################################

-- NOTE: mean / percentile values are calculated manually because of an issue with
-- empty points in the queries https://github.com/influxdata/influxdb/issues/6967
function driver:_performStatsQuery(stats_query, tstart, tend)
  local data = influx_query(self.url .. "/query?db=".. self.db .."&epoch=s", stats_query, self.username, self.password)

  if (data and data.series and data.series[1] and data.series[1].values[1]) then
    local data_stats = data.series[1].values[1]
    local total = data_stats[2]

    return {total=total, average = (total / (tend - tstart))}
  end

  return nil
end

-- ##############################################

local function getDatabaseName(schema, db)
  return ternary(schema.options.influx_internal_query, '_internal', db)
end

function driver:_makeSeriesQuery(query_schema, metrics, tags, tstart, tend, time_step, schema)
  local internal_query = schema.options.influx_internal_query

  if internal_query ~= nil then
    return internal_query(self, schema, tstart, tend, time_step)
  end

  return 'SELECT '.. table.concat(metrics, ",") ..' FROM ' .. query_schema .. where_tags(tags) ..
      " time >= " .. tstart .. "000000000 AND time <= " .. tend .. "000000000" ..
      " GROUP BY TIME(".. time_step .."s)"
end

function driver:query(schema, tstart, tend, tags, options)
  local metrics = {}
  local retention_policy = getSchemaRetentionPolicy(schema, tstart, tend, options)
  local query_schema, raw_step, data_type = retentionPolicyToSchema(schema, retention_policy, self.db)

  tend = fixTendForRetention(tend, retention_policy)
  local time_step = ts_common.calculateSampledTimeStep(raw_step, tstart, tend, options)

  -- NOTE: this offset is necessary to fix graph edge points when data insertion is not aligned with tstep
  local unaligned_offset = raw_step - 1

  for i, metric in ipairs(schema._metrics) do
    -- NOTE: why we need to device by time_step ? is MEAN+GROUP BY TIME bugged?
    if data_type == ts_common.metrics.counter then
      metrics[i] = "(DERIVATIVE(MEAN(\"" .. metric .. "\")) / ".. time_step ..") as " .. metric
    else -- gauge / derivative
      metrics[i] = schema:getAggregationFunction() .. "(\"".. metric .."\") as " .. metric
    end
  end

  -- NOTE: GROUP BY TIME and FILL do not work well together! Additional zeroes produce non-existent derivative values
  -- Will perform fill manually below
  --[[
  SELECT (DERIVATIVE(MEAN("bytes")) / 60) as bytes
    FROM "iface:ndpi" WHERE protocol='SSL' AND ifid='2'
    AND time >= 1531991910000000000 AND time <= 1532002710000000000
    GROUP BY TIME(60s)
  ]]
  local query = self:_makeSeriesQuery(query_schema, metrics, tags, tstart, tend + unaligned_offset, time_step, schema)

  local url = self.url
  local data = {}
  local series, count

  -- Perform an additional query to determine the last point in the raw data
  local last_ts_query = getLastTsQuery(schema, query_schema, tags)

  local jres = influx_query_multi(url .. "/query?db=".. getDatabaseName(schema, self.db) .."&epoch=s", string.format("%s;%s", query, last_ts_query), self.username, self.password, options)
  local last_ts = os.time()

  if(jres and jres.results and (#jres.results == 2)) then
    if jres.results[1].series then
      data = jres.results[1]
    end

    if jres.results[2].series and jres.results[2].series[1].values then
      last_ts = jres.results[2].series[1].values[1][1]
    end
  end

  if table.empty(data) then
    series, count = ts_common.fillSeries(schema, tstart + time_step, tend, time_step, options.fill_value)
  else
    -- Note: we are working with intervals because of derivatives. The first interval ends at tstart + time_step
    -- which is the first value returned by InfluxDB
    series, count, tstart = influx2Series(schema, tstart + time_step, tend, tags, options, data.series[1], time_step, last_ts)
  end

  local total_serie = nil
  local stats = nil

  if options.calculate_stats then
    local is_single_serie = (#series == 1)

    if is_single_serie then
      -- optimization
      -- table.clone needed as total_serie can be modified below (table.insert)
      total_serie = table.clone(series[1].data)
    else
      -- try to inherit label from existing series
      local label = series and series[1] and series[1].label
      total_serie = self:_makeTotalSerie(schema, query_schema, raw_step, tstart + time_step, tend, tags, options, url, time_step, label, unaligned_offset, data_type)
    end

    if total_serie then
      stats = ts_common.calculateStatistics(total_serie, time_step, tend - tstart, data_type)

      stats = stats or {}
      stats.by_serie = {}

      for k, v in pairs(series) do
        local s = ts_common.calculateStatistics(v.data, time_step, tend - tstart, data_type)
        stats.by_serie[k] = s

        -- Remove the total for now as it requires a complex computation (see below)
        s.total = nil
      end

      if stats.total ~= nil then
        -- override total and average
        -- NOTE: using -1 to avoid overflowing into the next hour
        local stats_query = "(SELECT ".. table.concat(schema._metrics, " + ") .. ' AS value FROM ' .. query_schema ..
          ' ' ..getWhereClause(tags, tstart, tend, -1) .. ")"
        if data_type == ts_common.metrics.counter then
          stats_query = "(SELECT NON_NEGATIVE_DIFFERENCE(value) as value FROM " .. stats_query .. ")"
        end
        stats_query = "SELECT SUM(value) as value FROM " .. stats_query

        if data_type == ts_common.metrics.derivative then
          stats_query = "SELECT (value * ".. time_step ..") as value FROM (" .. stats_query .. ")"
        end

        stats = table.merge(stats, self:_performStatsQuery(stats_query, tstart, tend))
      end
    end
  end

  if options.initial_point then
    local initial_metrics = {}

    for idx, metric in ipairs(schema._metrics) do
      initial_metrics[idx] = "FIRST(" .. metric .. ")"
    end

    local query = self:_makeSeriesQuery(query_schema, metrics, tags, tstart-time_step, tstart+unaligned_offset, time_step, schema)
    local data = influx_query(url .. "/query?db=".. getDatabaseName(schema, self.db) .."&epoch=s", query, self.username, self.password, options)

    if table.empty(data) then
      -- Data fill
      for _, serie in pairs(series) do
        table.insert(serie.data, 1, options.fill_value)
      end
    else
      local values = data.series[1].values[1]
      for i=2, #values do
        local max_val = ts_common.getMaxPointValue(schema, series[i-1].label, tags)
        local val = ts_common.normalizeVal(values[i], max_val, options)
        table.insert(series[i-1].data, 1, val)
      end
    end

    -- shift tstart as we added one point
    tstart = tstart - time_step

    if total_serie then
      local label = series and series[1].label
      local additional_pt = self:_makeTotalSerie(schema, query_schema, raw_step, tstart-time_step, tstart, tags, options, url, time_step, label, unaligned_offset, data_type) or {options.fill_value}
      table.insert(total_serie, 1, additional_pt[1])
    end

    count = count + 1
  end

  if options.calculate_stats and total_serie then
    stats = table.merge(stats, ts_common.calculateMinMax(total_serie))
  end

  local rv = {
    start = tstart,
    step = time_step,
    raw_step = raw_step,
    count = count,
    series = series,
    statistics = stats,
    source_aggregation = retention_policy or "raw",
    additional_series = {
      total = total_serie,
    },
  }

  return rv
end

-- ##############################################

function driver:_exportErrorMsg(ret)
  local err_msg = i18n("delete_data.msg_err_unknown")
  local suffix = ""

  if ret ~= nil then
    if ret.error_msg ~= nil then
      err_msg = ret.error_msg .. "."
    elseif ret.CONTENT ~= nil then
      local content = json.decode(ret.CONTENT)

      if((content ~= nil) and (content.error ~= nil)) then
        -- In case multiple lines are returned, only fetch the first one
        local errors = string.split(content.error, "\n") or {content.error}
        err_msg = errors[1] or content.error

        if string.find(err_msg, "max-values-per-tag limit exceeded", nil, true) ~= nil then
          suffix = ". " .. i18n("alert_messages.influxdb_partial_write")
        end
      elseif ret.RESPONSE_CODE ~= nil then
        err_msg = err_msg .. " [" .. ret.RESPONSE_CODE .. "]"
      end
    elseif ret.RESPONSE_CODE ~= nil then
        err_msg = err_msg .. " [" .. ret.RESPONSE_CODE .. "]"
    end
  end

  return i18n("alert_messages.influxdb_write_error", {influxdb=self.url, err=err_msg}) .. suffix
end

-- ##############################################

-- Exports a timeseries file in line format to InfluxDB
-- Returns a tuple(success, file_still_existing)
function driver:_exportTsFile(exportable)
  local rv = true
  local fname = exportable["fname"]

  if not ntop.exists(fname) then
    traceError(TRACE_ERROR, TRACE_CONSOLE,
	       string.format("Cannot find ts file %s. Some timeseries data will be lost.", fname))
    -- The file isn't in place, probably has been manually deleted or a write failure prevented
    -- it from being written. It's safe to remove the corresponding item from the queue as this
    -- won't be a recoverable export.
    ntop.lremCache(INFLUX_EXPORT_QUEUE, exportable["item"])
    return false
  end

  -- Delete the file after POST
  local delete_file_after_post = false
  local ret = ntop.postHTTPTextFile(self.username, self.password, self.url .. "/write?precision=s&db=" .. self.db, fname, delete_file_after_post, 30 --[[ timeout ]])

  if((ret == nil) or ((ret.RESPONSE_CODE ~= 200) and (ret.RESPONSE_CODE ~= 204))) then
    local msg = self:_exportErrorMsg(ret)
    ntop.setCache("ntopng.cache.influxdb.last_error", msg)

    rv = false
  else
    -- Clear last error
    ntop.delCache("ntopng.cache.influxdb.last_error")
  end

  return rv
end

-- ##############################################

-- Returns an indication of the current InfluxDB health.
-- Health is "green" when everything is working as expected,
-- "yellow" when there are recoverable issues, or "red" when
-- there is some critical error.
-- Health corresponds to the current status, i.e., a past
-- error, will no longer be considered
function driver:get_health()
   if is_dropping_points() then
      return "red"
   elseif is_failing_exports() then
      return "yellow"
   end

   return "green"
end

-- ##############################################

local function deleteExportableFile(exportable)
   if exportable and exportable["fname"] then
      if ntop.exists(exportable["fname"]) then
	 os.remove(exportable["fname"])
	 traceError(TRACE_INFO, TRACE_CONSOLE, "Removed file "..exportable["fname"])
      end
   end
end

-- ##############################################

-- When we giveup for a certain exportable, that is, when we are not
-- going to try and export it again, we call this function
local function dropExportable(exportable)
   inc_dropped_points(exportable["num_points"])
   deleteExportableFile(exportable)
end

-- ##############################################

function driver:_droppedPointsAlert()
   local alert_periodicity = 60
   local k = "ntopng.cache.influxdb_dropped_points_alert_triggered"

   if ntop.getCache(k) ~= "1" then
      local alerts_api = require("alerts_api")
      local alert_consts = require "alert_consts"

      alerts_api.store(
        alerts_api.influxdbEntity(self.url),
        alert_consts.alert_types.alert_influxdb_export_failure.create(
	   alert_consts.alert_severities.error,
	   alert_consts.alerts_granularities.min,
	   self.url
	)
      )

      -- Just to avoid doing :trigger too often
      ntop.setCache(k, "1", alert_periodicity / 2)
   end
end

-- ##############################################

-- Call this function when an exportable has been sent to InfluxDB
-- with success
local function exportableSuccess(exportable)
   inc_exported_points(exportable["num_points"])
   inc_exports()
   deleteExportableFile(exportable)
end

-- ##############################################

-- Call this function when the export has failed but it is going
-- to be tried again
local function exportableFailure(exportable)
   inc_failed_exports()
end

-- ##############################################

function driver:_performExport(exportable)
   local time_key = "ntopng.cache.influxdb_export_time_" .. self.db .. "_" .. exportable["ifid"]
   local prev_t = tonumber(ntop.getCache(time_key)) or 0

   local start_t = ntop.gettimemsec()
   local rv = self:_exportTsFile(exportable)
   local end_t = ntop.gettimemsec()

   if rv then
      -- Successfully exported
      traceError(TRACE_INFO, TRACE_CONSOLE,
		 string.format("Successfully exported %u points in %.2fs", exportable["num_points"], (end_t - start_t)))
      ntop.setCache(time_key, tostring(math.max(prev_t, exportable["time_ref"])))
   end

   return rv
end

-- ##############################################

local function getExportable(item)
   local parts = split(item, "|")
   local ifid_str = parts[1]
   local ifid = tonumber(ifid_str)
   local time_ref = tonumber(parts[2])
   local export_id = tonumber(parts[3])
   local num_points = tonumber(parts[4])
   local res = {item = item}

   if not ifid or not time_ref or not export_id or not num_points then
      traceError(TRACE_ERROR, TRACE_CONSOLE, "Missing mandatory data from the exportable item "..(item or ''))
      return res
   end

   res["item"]       = item
   res["fname"]      = os_utils.fixPath(dirs.workingdir .. "/" .. ifid .. "/ts_export/" .. export_id .. "_" .. time_ref)
   res["ifid_str"]   = ifid_str
   res["ifid"]       = ifid
   res["num_points"] = num_points
   res["time_ref"]   = time_ref

   return res
end

-- ##############################################

function driver:export()
   interface.select(getSystemInterfaceId())

   local num_ifaces = table.len(interface.getIfNames())
   local high_value = (INFLUX_MAX_EXPORT_QUEUE_LEN_HIGH * num_ifaces)
   local low_value = (INFLUX_MAX_EXPORT_QUEUE_LEN_LOW * num_ifaces)
   local max_value = (INFLUX_MAX_EXPORT_QUEUE_TRIM_LEN * num_ifaces)
   local num_pending = ntop.llenCache(INFLUX_EXPORT_QUEUE)

   traceError(TRACE_INFO, TRACE_CONSOLE, "Exporting "..num_pending.." items")

   if(num_pending >= high_value) then
      -- Export is blocked. Inhibit subsequent writes.
      setExportQueueFull(true)
   elseif(num_pending <= low_value) then
      -- The problem was solved
      setExportQueueFull(false)
   end

   if num_pending == 0 then
      return
   end

   while(num_pending > max_value) do
      -- NOTE: this should not happen, but if it does we manually trim the queue
      local being_dropped = ntop.lpopCache(INFLUX_EXPORT_QUEUE)
      local exportable_to_be_dropped = getExportable(being_dropped)

      dropExportable(exportable_to_be_dropped)

      traceError(TRACE_INFO, TRACE_CONSOLE, "Dropped old item "..(being_dropped or ''))

      num_pending = num_pending - 1
   end

   if(ntop.getCache(INFLUX_FLAG_IS_CURRENTLY_DROPPING) == "true") then
      self:_droppedPointsAlert()
      ntop.delCache(INFLUX_FLAG_IS_CURRENTLY_DROPPING)
   end

   -- Post the guys using a pretty long timeout
   local pending_exports = ntop.lrangeCache(INFLUX_EXPORT_QUEUE, 0, -1)

   if not pending_exports then
      return
   end

   -- Process pending exports older-to-newer
   for _, cur_export in ipairs(pending_exports) do
      local exportable = getExportable(cur_export)
      local res = self:_performExport(exportable)

      if res then
	 -- export SUCCEDED
	 exportableSuccess(exportable)
	 ntop.lremCache(INFLUX_EXPORT_QUEUE, cur_export)
      else
	 -- export FAILED, retry next time
	 exportableFailure(exportable)
      end
   end

   interface.select(getSystemInterfaceId())
end

-- ##############################################

function driver:getLatestTimestamp(ifid)
  local k = "ntopng.cache.influxdb_export_time_" .. self.db .. "_" .. ifid
  local v = tonumber(ntop.getCache(k))

  if v ~= nil then
    return v
  end

  return os.time()
end

-- ##############################################

-- At least 2 values are needed otherwise derivative will return empty
local min_values_list_series = 2

local function makeListSeriesQuery(schema, tags_filter, wildcard_tags, start_time)
  -- NOTE: do not use getQuerySchema here, otherwise we'll miss series

  -- NOTE: time based query not currently supported on show tags/series, using select
  -- https://github.com/influxdata/influxdb/issues/5668
  --[[
  SELECT * FROM "iface:ndpi_categories"
    WHERE ifid='2' AND time >= 1531981349000000000
    GROUP BY category
    LIMIT 2
  ]]
  return 'SELECT * FROM "' .. schema.name .. '"' .. where_tags(tags_filter) ..
      " time >= " .. start_time .. "000000000" ..
      ternary(not table.empty(wildcard_tags), " GROUP BY " .. table.concat(wildcard_tags, ","), "") ..
      " LIMIT " .. min_values_list_series
end

local function processListSeriesResult(data, schema, tags_filter, wildcard_tags)
  if table.empty(data) then
    return nil
  end

  if table.empty(data.series) then
    return nil
  end

  if table.empty(wildcard_tags) then
    -- Simple "exists" check
    if #data.series[1].values >= min_values_list_series then
      return {tags_filter}
    else
      return nil
    end
  end

  local res = {}

  for _, serie in pairs(data.series) do
    if #serie.values < min_values_list_series then
      goto continue
    end

    for _, value in pairs(serie.values) do
      local tags = {}

      for i=2, #value do
        local tag = serie.columns[i]

        -- exclude metrics
        if schema.tags[tag] ~= nil then
          tags[tag] = value[i]
        end
      end

      for key, val in pairs(serie.tags) do
        tags[key] = val
      end

      res[#res + 1] = tags
      break
    end

    ::continue::
  end

  return res
end

-- ##############################################

function driver:listSeries(schema, tags_filter, wildcard_tags, start_time)
  local query = makeListSeriesQuery(schema, tags_filter, wildcard_tags, start_time)
  local url = self.url
  local data = influx_query(url .. "/query?db=".. self.db, query, self.username, self.password)

  return processListSeriesResult(data, schema, tags_filter, wildcard_tags)
end

-- ##############################################

function driver:exists(schema, tags_filter, wildcard_tags)
  if schema.options.influx_internal_query then
    -- internal metrics always exist
    return(true)
  end

  -- Ignore wildcard_tags to avoid exessive points returned due to group by
  wildcard_tags = {}

  local query = makeListSeriesQuery(schema, tags_filter, wildcard_tags, 0)
  local url = self.url
  local data = influx_query(url .. "/query?db=".. self.db, query, self.username, self.password)

  return(not table.empty(processListSeriesResult(data, schema, tags_filter, wildcard_tags)))
end

-- ##############################################

function driver:listSeriesBatched(batch)
  local max_batch_size = 30
  local url = self.url
  local rv = {}
  local idx_to_batchid = {}
  local internal_results = {}

  for i=1,#batch,max_batch_size do
    local queries = {}

    -- Prepare the batch
    for j=i,math.min(i+max_batch_size-1, #batch) do
      local cur_query = batch[j]

      if cur_query.schema.options.influx_internal_query then
        -- internal metrics always exist
        internal_results[j] = {{}} -- exists
      else
        local idx = #queries +1
        idx_to_batchid[idx] = j
        queries[idx] = makeListSeriesQuery(cur_query.schema, cur_query.filter_tags, cur_query.wildcard_tags, cur_query.start_time)
      end
    end

    local query_str = table.concat(queries, ";")
    local data = influx_query_multi(url .. "/query?db=".. self.db, query_str, self.username, self.password)

    -- Collect the results
    if data and data.results then
      for idx, result in pairs(data.results) do
        local j = idx_to_batchid[idx]
        local cur_query = batch[j]
        local result = processListSeriesResult(result, cur_query.schema, cur_query.filter_tags, cur_query.wildcard_tags)
        rv[j] = result
      end
    end
  end

  return table.merge(rv, internal_results)
end

-- ##############################################

function getWhereClause(tags, tstart, tend, unaligned_offset)
  return where_tags(tags) .. ' time >= '.. tstart ..'000000000 AND time <= '.. (tend + unaligned_offset) .. "000000000"
end

function driver:topk(schema, tags, tstart, tend, options, top_tags)
  if #top_tags ~= 1 then
    traceError(TRACE_ERROR, TRACE_CONSOLE, "InfluxDB driver expects exactly one top tag, " .. #top_tags .. " found")
    return nil
  end

  local top_tag = top_tags[1]
  local retention_policy = getSchemaRetentionPolicy(schema, tstart, tend, options)
  local query_schema, raw_step, data_type = retentionPolicyToSchema(schema, retention_policy, self.db)
  tend = fixTendForRetention(tend, retention_policy)

  -- NOTE: this offset is necessary to fix graph edge points when data insertion is not aligned with tstep
  local unaligned_offset = raw_step - 1

  local derivate_metrics = {}
  local sum_metrics = {}
  local all_metrics = table.concat(schema._metrics, ", ")

  for idx, metric in ipairs(schema._metrics) do
    if data_type == "counter" then
      derivate_metrics[idx] = 'NON_NEGATIVE_DIFFERENCE('.. metric .. ') as ' .. metric
    else -- derivative
      derivate_metrics[idx] = '('.. metric .. ' * '.. raw_step ..') as ' .. metric
    end
    sum_metrics[idx] = 'SUM('.. metric .. ') as ' .. metric
  end

  --[[
    SELECT TOP(value,protocol,10) FROM
      (SELECT SUM(value) AS value FROM
        (SELECT NON_NEGATIVE_DIFFERENCE(value) as value FROM
          (SELECT protocol, (bytes_sent + bytes_rcvd) AS "value" FROM "host:ndpi" WHERE host='192.168.1.1'
            AND ifid='1' AND time >= 1537540320000000000 AND time <= 1537540649000000000)
          GROUP BY protocol)
        GROUP BY protocol)
  ]]
  -- Aggregate into 1 metric and filter
  local base_query = '(SELECT '.. top_tag ..', (' .. table.concat(schema._metrics, " + ") ..') AS "value", '..
      all_metrics .. ' FROM '.. query_schema ..
      ' '.. getWhereClause(tags, tstart, tend, -1) ..')'

   -- Calculate difference between counter values
  if data_type == "counter" then
    base_query = '(SELECT NON_NEGATIVE_DIFFERENCE(value) as value, '.. table.concat(derivate_metrics, ", ")  ..
      ' FROM ' .. base_query .. " GROUP BY ".. top_tag ..")"
  else
    -- derivative
    base_query = '(SELECT (value * '.. raw_step ..') as value, '.. table.concat(derivate_metrics, ", ")  ..
      ' FROM ' .. base_query .. " GROUP BY ".. top_tag ..")"
  end

  -- Sum the traffic
  base_query = '(SELECT SUM(value) AS value, '.. table.concat(sum_metrics, ", ")  ..
      ' FROM '.. base_query .. ' GROUP BY '.. top_tag ..')'

  -- Calculate TOPk
  local query = 'SELECT TOP(value,'.. top_tag ..','.. options.top ..'), '.. all_metrics ..' FROM ' .. base_query

  local url = self.url
  local data = influx_query(url .. "/query?db=".. self.db .."&epoch=s", query, self.username, self.password, options)

  if table.empty(data) then
    return data
  end

  if table.empty(data.series) then
    return {}
  end

  data = data.series[1]

  local res = {}

  for idx, value in pairs(data.values) do
    -- top value
    res[idx] = value[2]
  end

  local sorted = {}

  for idx in pairsByValues(res, rev) do
    local value = data.values[idx]

    if value[2] > 0 then
      local partials = {}

      for idx=4, #value do
        partials[data.columns[idx]] = value[idx]
      end

      sorted[#sorted + 1] = {
        tags = table.merge(tags, {[top_tag] = value[3]}),
        value = value[2],
        partials = partials,
      }
    end
  end

  local time_step = ts_common.calculateSampledTimeStep(raw_step, tstart, tend, options)
  local label = series and series[1].label
  local total_serie = self:_makeTotalSerie(schema, query_schema, raw_step, tstart, tend, tags, options, url, time_step, label, unaligned_offset, data_type)
  local stats = nil

  if options.calculate_stats and total_serie then
    stats = ts_common.calculateStatistics(total_serie, time_step, tend - tstart, data_type)

    if stats.total ~= nil then
      -- override total and average
      -- NOTE: sum must be calculated on individual top_tag fields to avoid
      -- calculating DIFFERENCE on a decreasing serie (e.g. this can happend on a top hosts query
      -- when a previously seen host becomes idle, which causes its traffic contribution on the total to become zero on next points)
      local stats_query = "SELECT SUM(value) as value FROM " .. base_query
      stats = table.merge(stats, self:_performStatsQuery(stats_query, tstart, tend))
    end
  end

  if options.initial_point and total_serie then
    local additional_pt = self:_makeTotalSerie(schema, query_schema, raw_step, tstart-time_step, tstart, tags, options, url, time_step, label, unaligned_offset, data_type) or {options.fill_value}
    table.insert(total_serie, 1, additional_pt[1])

    -- shift tstart as we added one point
    tstart = tstart - time_step
  end

  if options.calculate_stats and total_serie then
    stats = table.merge(stats, ts_common.calculateMinMax(total_serie))
  end

  return {
    topk = sorted,
    statistics = stats,
    source_aggregation = retention_policy or "raw",
    additional_series = {
      total = total_serie,
    },
  }
end

-- ##############################################

function driver:queryTotal(schema, tstart, tend, tags, options)
  local query_schema, raw_step, data_type = getQuerySchema(schema, tstart, tend, self.db, options)
  local query

  if data_type == ts_common.metrics.counter then
    local metrics = {}
    local sum_metrics = {}

    for i, metric in ipairs(schema._metrics) do
      metrics[i] = "NON_NEGATIVE_DIFFERENCE(" .. metric .. ") as " .. metric
      sum_metrics[i] = "SUM(" .. metric .. ") as " .. metric
    end

    -- SELECT SUM("bytes_sent") as "bytes_sent", SUM("bytes_rcvd") as "bytes_rcvd" FROM
    --  (SELECT DIFFERENCE("bytes_sent") AS "bytes_sent", DIFFERENCE("bytes_rcvd") AS "bytes_rcvd"
    --    FROM "host:traffic" WHERE ifid='1' AND host='192.168.1.1' AND time >= 1536321770000000000 AND time <= 1536322070000000000)
    query = 'SELECT ' .. table.concat(sum_metrics, ", ") .. ' FROM ' ..
    '(SELECT ' .. table.concat(metrics, ", ") .. ' FROM '.. query_schema .. where_tags(tags) ..
    ' time >= ' .. tstart .. '000000000 AND time <= ' .. tend .. '000000000)'
  else
    -- gauge/derivative
    local metrics = {}

    for i, metric in ipairs(schema._metrics) do
      metrics[i] = "(SUM(" .. metric .. ")"

      if data_type == "derivative" then
        metrics[i] = metrics[i] .. " * " .. raw_step
      end

      metrics[i] = metrics[i] .. ") as " .. metric
    end

    query = 'SELECT ' .. table.concat(metrics, ", ") .. ' FROM ' .. query_schema .. where_tags(tags) ..
      ' time >= ' .. tstart .. '000000000 AND time <= ' .. tend .. '000000000'
  end

  local url = self.url
  local data = influx_query(url .. "/query?db=".. self.db .."&epoch=s", query, self.username, self.password, options)

  if table.empty(data) then
    return data
  end

  data = data.series[1]
  local res = {}

  for i=2, #data.columns do
    local metric = data.columns[i]
    local total = data.values[1][i]

    res[metric] = total
  end

  return res
end

-- ##############################################

function driver:queryMean(schema, tags, tstart, tend, options)
  local metrics = {}
  local query_schema = getQuerySchema(schema, tstart, tend, self.db, options)

  for i, metric in ipairs(schema._metrics) do
    metrics[i] = "MEAN(" .. metric .. ") as " .. metric
  end

  local query = 'SELECT ' .. table.concat(metrics, ", ") .. ' FROM '.. query_schema .. where_tags(tags) ..
    ' time >= ' .. tstart .. '000000000 AND time <= ' .. tend .. '000000000'

  local url = self.url
  local data = influx_query(url .. "/query?db=".. self.db .."&epoch=s", query, self.username, self.password)

  if table.empty(data) then
    return data
  end

  data = data.series[1]
  local res = {}

  for i=2, #data.columns do
    local metric = data.columns[i]
    local average = data.values[1][i]

    res[metric] = average
  end

  return res
end

-- ##############################################

local function getInfluxdbVersion(url, username, password)
  local res = ntop.httpGet(url .. "/ping", username, password, getInfluxDBQueryTimeout(), true)
  if not res or ((res.RESPONSE_CODE ~= 200) and (res.RESPONSE_CODE ~= 204)) then
    local err_info = getResponseError(res)
    if err_info == 0 then
      err_info = i18n("prefs.is_influxdb_running")
    end
    local err = i18n("prefs.could_not_contact_influxdb", {msg=err_info})

    traceError(TRACE_ERROR, TRACE_CONSOLE, err)
    return nil, err
  end

  local content = res.CONTENT or ""
  -- case-insensitive match as HAProxy transforms headers to lowercase (see #3964)
  return string.match(content:lower(), "\nx%-influxdb%-version: v?([%d|%.]+)")
end

function driver:getInfluxdbVersion()
  return getInfluxdbVersion(self.url, self.username, self.password)
end

-- ##############################################

local function single_query(base_url, query, username, password)
  local data = influx_query(base_url, query, username, password)

  if data and data.series and data.series[1] and data.series[1].values[1] then
    return data.series[1].values[1][2]
  end

  return nil
end

-- ##############################################

function driver:getDiskUsage()
   local query = 'SELECT SUM(last) FROM (select LAST(diskBytes) FROM "monitor"."shard" where "database" = \''.. self.db ..'\' group by id)'
  return single_query(self.url .. "/query?db=_internal", query, self.username, self.password)
end

-- ##############################################

function driver:getMemoryUsage()
  --[[
     This function attempts to match the memory used by the process, memory which is
     the top/htop RSS (Resident Stack Size) which is what it actually matters.

     InfluxDB docs leak explanations of how to interpred memory-related numbers:
     https://docs.influxdata.com/platform/monitoring/influxdata-platform/tools/measurements-internal/#runtime

     So the formula below has been obtained by tentatives and it seems to be pretty
     close to what top reports.
  --]]
  local query = 'SELECT LAST(Sys) - LAST(HeapReleased) FROM "_internal".."runtime"'
  return single_query(self.url .. "/query?db=_internal", query, self.username, self.password)
end

-- ##############################################

function driver:getSeriesCardinality()
  local query = 'SELECT LAST("numSeries") FROM "_internal".."database" where "database"=\'' .. self.db ..'\''
  return single_query(self.url .. "/query?db=_internal", query, self.username, self.password)
end

-- ##############################################

function driver.getShardGroupDuration(days_retention)
  -- https://docs.influxdata.com/influxdb/v1.7/query_language/database_management/#description-of-syntax-1
  if days_retention < 2 then
    return "1h"
  elseif days_retention < 180 then
    return "1d"
  else
    return "7d"
  end
end

local function makeRetentionPolicyQuery(statement, rpname, dbname, retention)
  return(string.format('%s RETENTION POLICY "%s" ON "%s" DURATION %ud REPLICATION 1 SHARD DURATION %s',
    statement, rpname, dbname,
    retention, driver.getShardGroupDuration(retention)
  ))
end

-- ##############################################

local function updateCQRetentionPolicies(dbname, url, username, password)
  local query = string.format("SHOW RETENTION POLICIES ON %s", dbname)
  local res = influx_query(url .. "/query?db=".. dbname, query, username, password)
  local rp_1h_statement = "CREATE"
  local rp_1d_statement = "CREATE"

  if res and res.series and res.series[1] then
    for _, rp in pairs(res.series[1].values) do
      local rp_name = rp[1]

      if rp_name == "1h" then
        rp_1h_statement = "ALTER"
      elseif rp_name == "1d" then
        rp_1d_statement = "ALTER"
      end
    end
  end

  local queries = {
    makeRetentionPolicyQuery(rp_1d_statement, "1d", dbname, get1dDatabaseRetentionDays())
  }

  if(HOURLY_CQ_ENABLED) then
    queries[#queries + 1] = makeRetentionPolicyQuery(rp_1h_statement, "1h", dbname, get1hDatabaseRetentionDays())
  else
    -- Delete existing 1h aggregated data
    queries[#queries + 1] = "DROP RETENTION POLICY \"1h\" ON \"" .. dbname .. "\""
  end

  return multiQueryPost(queries, url, username, password)
end

-- ##############################################

local function toVersion(version_str)
  local parts = string.split(version_str, "%.")

  if (#parts ~= 3) then
    return nil
  end

  return {
    major = tonumber(parts[1]) or 0,
    minor = tonumber(parts[2]) or 0,
    patch = tonumber(parts[3]) or 0,
  }
end

local function isCompatibleVersion(version)
  local current = toVersion(version)
  local required = toVersion(MIN_INFLUXDB_SUPPORTED_VERSION)

  if (version == nil) or (required == nil) then
    return false
  end

  return (current.major == required.major) and
    ((current.minor > required.minor) or
      ((current.minor == required.minor) and (current.patch >= required.patch)))
end

function driver.init(dbname, url, days_retention, username, password, verbose)
  local timeout = getInfluxDBQueryTimeout()

  -- Check version
  if verbose then traceError(TRACE_NORMAL, TRACE_CONSOLE, "Contacting influxdb at " .. url .. " ...") end

  local version, err = getInfluxdbVersion(url, username, password)

  if((not version) and (err ~= nil)) then
    ntop.setCache("ntopng.cache.influxdb.last_error", err)
    return false, err
  elseif((not version) or (not isCompatibleVersion(version))) then
    local err = i18n("prefs.incompatible_influxdb_version_found",
      {required=MIN_INFLUXDB_SUPPORTED_VERSION, found=version, url="https://portal.influxdata.com/downloads"})

    ntop.setCache("ntopng.cache.influxdb.last_error", err)
    traceError(TRACE_ERROR, TRACE_CONSOLE, err)
    return false, err
  end

  -- Check existing database (this is used to prevent db creationg error for unprivileged users)
  if verbose then traceError(TRACE_NORMAL, TRACE_CONSOLE, "Checking database " .. dbname .. " ...") end
  local query = "SHOW DATABASES"
  local res = ntop.httpPost(url .. "/query", "q=" .. query, username, password, timeout, true)
  local db_found = false

  if res and (res.RESPONSE_CODE == 200) and res.CONTENT then
    local reply = json.decode(res.CONTENT)

    if reply and reply.results and reply.results[1] and reply.results[1].series then
      local dbs = reply.results[1].series[1]

      if((dbs ~= nil) and (dbs.values ~= nil)) then
        for _, row in pairs(dbs.values) do
          local user_db = row[1]

          if user_db == dbname then
            db_found = true
            break
          end
        end
      end
    end
  end

  if not db_found then
    -- Create database
    if verbose then traceError(TRACE_NORMAL, TRACE_CONSOLE, "Creating database " .. dbname .. " ...") end
    local query = "CREATE DATABASE \"" .. dbname .. "\""

    local res = ntop.httpPost(url .. "/query", "q=" .. query, username, password, timeout, true)
    if not res or (res.RESPONSE_CODE ~= 200) then
      local err = i18n("prefs.influxdb_create_error", {db=dbname, msg=getResponseError(res)})

      ntop.setCache("ntopng.cache.influxdb.last_error", err)
      traceError(TRACE_ERROR, TRACE_CONSOLE, err)
      return false, err
    end
  end

  if not db_found or days_retention ~= nil then
    -- New database or config changed
    days_retention = days_retention or getDatabaseRetentionDays()

    -- Set retention
    if verbose then traceError(TRACE_NORMAL, TRACE_CONSOLE, "Setting retention for " .. dbname .. " ...") end
    local query = makeRetentionPolicyQuery("ALTER", "autogen", dbname, days_retention)

    local res = ntop.httpPost(url .. "/query", "q=" .. query, username, password, timeout, true)
    if not res or (res.RESPONSE_CODE ~= 200) then
      local warning = i18n("prefs.influxdb_retention_error", {db=dbname, msg=getResponseError(res)})

      traceError(TRACE_WARNING, TRACE_CONSOLE, warning)
      -- This is just a warning, we can proceed
      --return false, err
    end

    -- NOTE: updateCQRetentionPolicies will be called automatically as driver:setup is triggered after this
  end

  ntop.delCache("ntopng.cache.influxdb.last_error")
  return true, i18n("prefs.successfully_connected_influxdb", {db=dbname, version=version})
end

-- ##############################################

function driver:delete(schema_prefix, tags)
  local url = self.url
  local measurement_pattern = ternary(schema_prefix == "", '//', '/^'.. schema_prefix ..':/')
  local query = 'DELETE FROM '.. measurement_pattern
  if not table.empty(tags) then
    query = query .. ' WHERE ' .. table.tconcat(tags, "=", " AND ", nil, "'")
  end
  local full_url = url .. "/query?db=".. self.db .."&q=" .. urlencode(query)

  local res = ntop.httpGet(full_url, self.username, self.password, getInfluxDBQueryTimeout(), true)

  if not res or (res.RESPONSE_CODE ~= 200) then
    traceError(TRACE_ERROR, TRACE_CONSOLE, getResponseError(res))
    return false
  end

  return true
end

-- ##############################################

function driver:deleteOldData(ifid)
  -- NOTE: retention is perfomed automatically based on the database retention policy,
  -- no need for manual intervention
  return true
end

-- ##############################################

local function getCqQuery(dbname, tags, schema, source, dest, step, dest_step, resemple)
  local cq_name = string.format("%s__%s", schema.name, dest)
  local resemple_s = ""
  local _, _, data_type = retentionPolicyToSchema(schema, source, dbname)

  if resemple then
    resemple_s = "RESAMPLE FOR " .. resemple
  end

  if data_type == ts_common.metrics.counter then
    local sums = {}
    local diffs = {}

    for _, metric in ipairs(schema._metrics) do
      sums[#sums + 1] = string.format('(SUM(%s) / %u) as %s', metric, dest_step, metric)
      diffs[#diffs + 1] = string.format('NON_NEGATIVE_DIFFERENCE(%s) as %s', metric, metric)
    end

    sums = table.concat(sums, ",")
    diffs = table.concat(diffs, ",")

    --[[
    CREATE CONTINUOUS QUERY "iface:packets__1h" ON ntopng
    RESAMPLE FOR 2h
    BEGIN
      SELECT (SUM(packets)/3600) as packets
      INTO "1h"."iface:packets" FROM (
        SELECT NON_NEGATIVE_DIFFERENCE(packets) as packets
          FROM "autogen"."iface:packets"
      ) GROUP BY time(1h),ifid
    END]]
    return string.format([[
      CREATE CONTINUOUS QUERY "%s" ON %s
      %s
      BEGIN
        SELECT
          %s
          INTO "%s"."%s"
          FROM (
            SELECT
              %s
              FROM "%s"."%s"
          ) GROUP BY time(%s)%s%s
      END]], cq_name, dbname, resemple_s,
      sums, dest, schema.name,
      diffs, source, schema.name,
      dest, ternary(isEmptyString(tags), "", ","), tags)
  else
    local means = {}

    for _, metric in ipairs(schema._metrics) do
      means[#means + 1] = string.format('%s(%s) as %s', schema:getAggregationFunction(), metric, metric)
    end

    means = table.concat(means, ",")

    --[[
    CREATE CONTINUOUS QUERY "asn:rtt__1h" ON ntopng
    RESAMPLE FOR 2h
    BEGIN
      SELECT MEAN(millis_rtt) as millis_rtt
      INTO "1h"."asn:rtt" FROM (
        SELECT MEAN(millis_rtt) as millis_rtt
        FROM "autogen"."asn:rtt"
        GROUP BY time(300s),ifid,asn FILL(0)
      ) GROUP BY time(1h),ifid,asn
    END]]
    return string.format([[
      CREATE CONTINUOUS QUERY "%s" ON %s
      %s
      BEGIN
        SELECT
          %s
          INTO "%s"."%s"
          FROM (
            SELECT
              %s
              FROM "%s"."%s"
              GROUP BY time(%us)%s%s
              FILL(0)
          ) GROUP BY time(%s)%s%s
      END]], cq_name, dbname, resemple_s,
      means, dest, schema.name,
      means, source, schema.name,
      step, ternary(isEmptyString(tags), "", ","), tags,
      dest, ternary(isEmptyString(tags), "", ","), tags)
  end
end

function driver:setup(ts_utils)
  local queries = {}
  local max_batch_size = 25 -- note: each query is about 400 characters

  -- Clear saved values (e.g., number of exported points) as
  -- we want to start clean and keep values since-ntopng-startup
  del_all_vals()

  -- Ensure that the database exists
  driver.init(self.db, self.url, nil, self.username, self.password)

  if not updateCQRetentionPolicies(self.db, self.url, self.username, self.password) then
    traceError(TRACE_ERROR, TRACE_CONSOLE, "InfluxDB setup failed")
    return false
  end

  if not isRollupEnabled() then
    -- Nothing more to do
    return true
  end

  -- Continuos Queries stuff
  ts_utils.loadSchemas()
  local schemas = ts_utils.getLoadedSchemas()

  -- NOTE: continuos queries cannot be altered, so they must be manually
  -- dropped and created in case of changes
  for _, schema in pairs(ts_utils.getPossiblyChangedSchemas()) do
    if(HOURLY_CQ_ENABLED) then
      queries[#queries + 1] = 'DROP CONTINUOUS QUERY "'.. schema ..'__1h" ON ' .. self.db
    end

    queries[#queries + 1] = 'DROP CONTINUOUS QUERY "'.. schema ..'__1d" ON ' .. self.db
  end

  -- Needed to handle migration
  local previous_1h_enabled = not (ntop.getPref(HOURLY_CQ_DISABLED_KEY) == "1")
  local migration_necessary = (previous_1h_enabled ~= HOURLY_CQ_ENABLED)

  for _, schema in pairs(schemas) do
    local tags = table.concat(schema._tags, ",")

    if((#schema._metrics == 0) or (schema.options.influx_internal_query)) then
      goto continue
    end

    if(migration_necessary) then
        -- NOTE: dropping all the continuous queries all together is not possible
        -- as InfluxDB does not provide an API for this and calling "SHOW CONTINUOUS QUERIES"
        -- yelds too much result data
        queries[#queries + 1] = 'DROP CONTINUOUS QUERY "'.. schema.name ..'__1h" ON ' .. self.db
        queries[#queries + 1] = 'DROP CONTINUOUS QUERY "'.. schema.name ..'__1d" ON ' .. self.db
    end

    if(HOURLY_CQ_ENABLED) then
      local cq_1h = getCqQuery(self.db, tags, schema, "autogen", "1h", schema.options.step, 3600, "2h")
      local cq_1d = getCqQuery(self.db, tags, schema, "1h", "1d", 3600, 86400)

      queries[#queries + 1] = cq_1h:gsub("\n", ""):gsub("%s%s+", " ")
      queries[#queries + 1] = cq_1d:gsub("\n", ""):gsub("%s%s+", " ")
    else
      local cq_1d = getCqQuery(self.db, tags, schema, "autogen", "1d", schema.options.step, 86400)

      queries[#queries + 1] = cq_1d:gsub("\n", ""):gsub("%s%s+", " ")
    end

    if #queries >= max_batch_size then
      if not multiQueryPost(queries, self.url, self.username, self.password) then
        traceError(TRACE_ERROR, TRACE_CONSOLE, "InfluxDB setup failed")
        return false
      end
      queries = {}
    end

    ::continue::
  end

  if #queries > 0 then
    if not multiQueryPost(queries, self.url, self.username, self.password) then
      traceError(TRACE_ERROR, TRACE_CONSOLE, "InfluxDB setup() failed")
      return false
    end
  end

  if(migration_necessary) then
    ntop.setPref(HOURLY_CQ_DISABLED_KEY, ternary(HOURLY_CQ_ENABLED, "0", "1"))

    -- Need to recalculate it
    ntop.delCache(FIRST_AGGREGATION_TIME_KEY)

    traceError(TRACE_NORMAL, TRACE_CONSOLE, "InfluxDB CQ migration completed")
  end

  if tonumber(ntop.getPref(FIRST_AGGREGATION_TIME_KEY)) == nil then
    local first_rp = ternary(HOURLY_CQ_ENABLED, "1h", "1d")
    local res = influx_query(self.url .. "/query?db=".. self.db .."&epoch=s",
      'SELECT FIRST(bytes) FROM "'.. first_rp ..'"."iface:traffic"', self.username, self.password)
    local first_t = os.time()

    if res and res.series and res.series[1].values then
      local v = res.series[1].values[1][1]

      if v ~= nil then
        first_t = v
      end
    end

    ntop.setPref(FIRST_AGGREGATION_TIME_KEY, tostring(first_t))
  end

  return true
end

-- ##############################################

function driver.getExportQueueLength()
  return(ntop.llenCache(INFLUX_EXPORT_QUEUE))
end

-- ##############################################

return driver