promise/plugin/check_free_disk_space: add options

Thanks to the new options in the software release,
it is now possible to have a customizable threshold.
Moreover, we can display the 3 biggest partitions if the
remaining space is smaller than the threshold. Finally,
we can have a prediction of the remaining disk space in 10 days (by default).

slapos/promise/plugin/check_free_disk_space.py

@@ -11,8 +11,20 @@ import sqlite3
 import argparse
 import datetime
 import psutil
+import itertools
+import warnings
+import pkgutil
 
 from slapos.collect.db import Database
+from contextlib import closing
+
+# install pandas, numpy and statsmodels for ARIMA prediction
+try:
+  import pandas as pd
+  import numpy as np
+  from statsmodels.tsa.arima_model import ARIMA
+except ImportError:
+  pass
 
 @implementer(interface.IPromise)
 class RunPromise(GenericPromise):
@@ -22,10 +34,10 @@ class RunPromise(GenericPromise):
     # check disk space at least every 3 minutes
     self.setPeriodicity(minute=3)
 
-
-
   def getDiskSize(self, disk_partition, database):
     database = Database(database, create=False, timeout=10)
+    # by using contextlib.closing, we don't need to close the database explicitly
+    with closing(database):
       try:
         # fetch disk size
         database.connect()
@@ -44,15 +56,11 @@ class RunPromise(GenericPromise):
             not self.raiseOnDatabaseLocked(locked_message):
           return None
         raise
-    finally:
-      try:
-        database.close()
-      except Exception:
-        pass
     return disk_size
 
   def getFreeSpace(self, disk_partition, database, date, time):
     database = Database(database, create=False, timeout=10)
+    with closing(database):
       try:
         # fetch free disk space
         database.connect()
@@ -71,12 +79,143 @@ class RunPromise(GenericPromise):
             not self.raiseOnDatabaseLocked(locked_message):
           return 0
         raise
-    finally:
+    return int(disk_free)
+
+  def getBiggestPartitions(self, database, date, time):
+    # displays the 3 biggest partitions thanks to disk usage
+    limit = 3
+    database = Database(database, create=False, timeout=10)
+    with closing(database):
       try:
-        database.close()
+        database.connect()
+        date_time = date + ' ' + time
+        # gets the data recorded between the current date (date_time) and 24 hours earlier
+        where_query = "datetime(date || ' ' || time) >= datetime('%s', '-1 days') AND datetime(date || ' ' || time) <= datetime('%s')"
+        # gets only the most recent data for each partition
+        result = database.select(
+          "folder",
+          columns = "partition, disk_used*1024, max(datetime(date || ' ' || time))",
+          where =  where_query % (date_time, date_time),
+          group = "partition",
+          order = "disk_used DESC",
+          limit = limit).fetchall()
+        if not result or not result[0]:
+          self.logger.info("No result from collector database in table folder: skipped")
+          return None
+      except sqlite3.OperationalError as e:
+        # if database is still locked after timeout expiration (another process is using it)
+        # we print warning message and try the promise at next run until max warn count
+        locked_message = "database is locked"
+        if locked_message in str(e) and \
+            not self.raiseOnDatabaseLocked(locked_message):
+          return None
+        raise
+    return result
+
+  def evaluateArimaModel(self, X, arima_order):
+    """
+    Evaluate an ARIMA model for a given order (p,d,q) with the MSE which
+    measures the average of the squares of the errors.
+    """
+    # take 66% of the data for training and 33% for testing
+    train_size = int(len(X) * 0.66)
+    train, test = X[0:train_size], X[train_size:]
+    history = [x for x in train]
+    # make predictions
+    predictions = list()
+    for t in range(len(test)):
+      with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        model = ARIMA(history, order=arima_order)
+        model_fit = model.fit(disp=-1)
+        yhat = model_fit.forecast()[0]
+        predictions.append(yhat)
+        history.append(test[t])
+    # calculate out of sample error
+    rmse = (np.square(np.subtract(test.values, np.hstack(predictions))).mean())**0.5
+    return rmse
+
+  def evaluateModels(self, dataset, p_values, d_values, q_values):
+    """
+    Evaluate combinations of p, d and q values for an ARIMA model
+    """
+    dataset = dataset.astype('float32')
+    best_score, best_cfg = float("inf"), None
+    for p in p_values:
+      for d in d_values:
+        for q in q_values:
+          order = (p,d,q)
+          try:
+            rmse = self.evaluateArimaModel(dataset, order)
+            if rmse < best_score:
+              best_score, best_cfg = rmse, order
           except Exception:
             pass
-    return int(disk_free)
+    return best_cfg
+
+  def diskSpacePrediction(self, disk_partition, database, date, time, day_range):
+    """
+    Returns an estimation of free disk space left depending on
+    the day_range parameter.
+
+    It uses Arima in order to predict data thanks to the 15 days before.
+    """
+    database = Database(database, create=False, timeout=10)
+    with closing(database):
+      try:
+        database.connect()
+        # get one data per day, where each data is at the same time
+        where_query = "time between '%s:00' and '%s:30' and partition='%s'" % (
+          time, time, disk_partition)
+        result = database.select(
+          "disk",
+          columns = "free, datetime(date || ' ' || time)",
+          where = where_query,
+          order = "datetime(date || ' ' || time) ASC").fetchall()
+        # checks that there are at least 14 days of data
+        if (not result) or (len(result) < 14):
+          self.logger.info("No or not enough results from collector database in table disk: no prediction")
+          return None
+        # put the list in pandas dataframe format and set the right types
+        df = pd.DataFrame(data=result, columns=['free', 'date'])
+        df.loc[:,'date'] = pd.to_datetime(df.date)
+        df = df.astype({'free': np.float})
+        df = df.set_index('date')
+        # find the best configuration by trying different combinations
+        p_values = d_values = q_values = range(0, 3)
+        best_cfg = self.evaluateModels(df.free, p_values, d_values, q_values)
+        # set the days to be predicted
+        max_date_predicted = day_range+1
+        future_index_date = pd.date_range(df.index[-1], freq='24H', periods=max_date_predicted)
+        try:
+          # disabling warnings during the ARIMA calculation
+          with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            model_arima = ARIMA(df, order=best_cfg)
+            # disp < 0 means no output about convergence information
+            model_arima_fit = model_arima.fit(disp=-1)
+            # save ARIMA predictions
+            fcast, _, conf = model_arima_fit.forecast(max_date_predicted, alpha=0.05)
+          # pass the same index as the others
+          fcast = pd.Series(fcast, index=future_index_date)
+          if fcast.empty:
+            self.logger.info("Arima prediction: none. Skipped prediction")
+            return None
+        except Exception:
+          self.logger.info("Arima prediction error: skipped prediction")
+          return None
+        # get results with 95% confidence
+        lower_series = pd.Series(conf[:, 0], index=future_index_date)
+        upper_series = pd.Series(conf[:, 1], index=future_index_date)
+        return fcast, lower_series, upper_series
+      except sqlite3.OperationalError as e:
+        # if database is still locked after timeout expiration (another process is using it)
+        # we print warning message and try the promise at next run until max warn count
+        locked_message = "database is locked"
+        if locked_message in str(e) and \
+            not self.raiseOnDatabaseLocked(locked_message):
+          return None
+        raise
 
   def raiseOnDatabaseLocked(self, locked_message):
     max_warn = 10
@@ -149,7 +288,7 @@ class RunPromise(GenericPromise):
       disk_partition = self.getConfig('test-disk-partition', '/dev/sda1')
     else:
       # get last minute
-      now = datetime.datetime.now()
+      now = datetime.datetime.utcnow()
       currentdate = now.strftime('%Y-%m-%d')
       currenttime = now - datetime.timedelta(minutes=1)
       currenttime = currenttime.time().strftime('%H:%M')
@@ -158,8 +297,7 @@ class RunPromise(GenericPromise):
     default_threshold = None
     if disk_size is not None:
       default_threshold = round(disk_size/(1024*1024*1024) * 0.05, 2)
-    threshold = float(self.getConfig('threshold', default_threshold) or 2.0)
-    threshold_days = float(self.getConfig('threshold-days', '30'))
+    threshold = float(self.getConfig('threshold', default_threshold) or default_threshold)
 
     free_space = self.getFreeSpace(disk_partition, db_path, currentdate,
                                    currenttime)
@@ -170,12 +308,57 @@ class RunPromise(GenericPromise):
       if inode_usage:
         self.logger.error(inode_usage)
       else:
-        self.logger.info("Disk usage: OK")
+        self.logger.info("Current disk usage: OK")
+        # if the option is enabled and the current disk size is large enough,
+        # we check the predicted remaining disk space
+        display_prediction = bool(int(self.getConfig('display-prediction', 0) or 0))
+        self.logger.info("Enable to display disk space predictions: %s" % display_prediction)
+        if display_prediction:
+          # check that the libraries are installed from the slapos.toolbox extra requires
+          pandas_found = pkgutil.find_loader("pandas")
+          numpy_found = pkgutil.find_loader("numpy")
+          statsmodels_found = pkgutil.find_loader("statsmodels")
+          # if one module isn't installed
+          if pandas_found is None or numpy_found is None or statsmodels_found is None:
+            self.logger.warning("Trying to use statsmodels and pandas " \
+              "but at least one module is not installed. Prediction skipped.")
             return
+          nb_days_predicted = int(self.getConfig('nb-days-predicted', 10) or 10)
+          disk_space_prediction_tuple = self.diskSpacePrediction(
+            disk_partition, db_path, currentdate, currenttime, nb_days_predicted)
+          if disk_space_prediction_tuple is not None:
+            fcast, lower_series, upper_series = disk_space_prediction_tuple
+            space_left_predicted = fcast.iloc[-1]
+            last_date_predicted = datetime.datetime.strptime(str(fcast.index[-1]),
+                                                            "%Y-%m-%d %H:%M:%S")
+            delta_days = (last_date_predicted.date() - \
+              datetime.datetime.strptime(currentdate, "%Y-%m-%d").date()).days
+            self.logger.info("Prediction: there will be %.2f G left on %s (%s days)." % (
+              space_left_predicted/(1024*1024*1024), last_date_predicted, delta_days))
+            if space_left_predicted <= threshold*1024*1024*1024:
+              self.logger.warning("The free disk space will be too low. " \
+                                "(disk size: %.2f G, threshold: %s G)" % (
+                                  disk_size/(1024*1024*1024), threshold))
+      return
+
+    message = "Free disk space low: remaining %.2f G (disk size: %.0f G, threshold: %.0f G)." % (
+      free_space/(1024*1024*1024), disk_size/(1024*1024*1024), threshold)
 
-    free_space = round(free_space/(1024*1024*1024), 2)
-    self.logger.error('Free disk space low: remaining %s G (threshold: %s G)' % (
-      free_space, threshold))
+    display_partition = bool(int(self.getConfig('display-partition', 0) or 0))
+    self.logger.info("Enable to display the 3 biggest partitions: %s" % display_partition)
+    if display_partition:
+      # display the 3 partitions that have the most storage capacity on the disk
+      big_partitions = self.getBiggestPartitions(db_path, currentdate, currenttime)
+      if big_partitions is not None:
+        for partition in big_partitions:
+          user_name, size_partition, date_checked = partition
+          partition_id = self.getConfig('partition-id', 'slappart')
+          # get the name of each partition by adding the user's number to the general name of the partition
+          partition_name = ''.join(x for x in partition_id if not x.isdigit()) + ''.join(filter(str.isdigit, user_name))
+          message += " The partition %s uses %.2f G (date checked: %s)." % (
+            partition_name, size_partition/(1024*1024*1024), date_checked)
+    # display the final error message
+    self.logger.error(message)
 
   def test(self):
     return self._test(result_count=1, failure_amount=1)