Loop processing gradually slow down with arcpy script

[YasunariMorita]

I try to the process below.

1. spatial search tile raster index polygon with feature
2. rasterize feature
3. combine rasterized feature and MODIS raster and export combine raster attribute table (loop for tiles) 
4. table operation with pandas and arcpy

This process works well.However, process gradually slow down.

I try this process for 51 times and check time. All input features are same.

The result is below.Each time_1,time_2 time_3 is described below.

time_1:finish process 2(rasterize feature)

time_2:finish process 3(combine and export table)

time_3:finish process 4(table operation)

elapsed_time:finish 1 feature process(save output in gdb file) 

• first time

INFO:root:land_00244 1/51 START
INFO:root:time_1:2.9405651092529297[sec]
INFO:root:time_2:3.73823618888855[sec]
INFO:root:time_2:4.30132269859314[sec]
INFO:root:time_2:4.817508697509766[sec]
INFO:root:time_2:5.239825248718262[sec]
INFO:root:time_3:9.50986933708191[sec]
INFO:root:elapsed_time:10.557835817337036[sec]

• 51 time

INFO:root:land_00244_49 51/51 START
INFO:root:time_1:14.089014768600464[sec]
INFO:root:time_2:19.874730348587036[sec]
INFO:root:time_2:25.050605058670044[sec]
INFO:root:time_2:30.117022275924683[sec]
INFO:root:time_2:35.16779851913452[sec]
INFO:root:time_3:51.1801700592041[sec]
INFO:root:elapsed_time:53.90104866027832[sec]

Even though same feature, process time increased 5 times!

All processes slow down.Why?

My environment is below.

 OS:windows 10 pro

 ArcGIS:10.5.1

 ArcGIS pro:2.0

But, in windows 7 and ArcGIS(10.4.1 and pro 1.4), this slow down didn't occur.

Thanks in advance!

from __future__ import print_function, unicode_literals, absolute_import
try:
import urllib2 # Python 2
except ImportError:
import urllib.request as urllib2 # Python 3
# Import arcpy module
import pandas as pd
import arcpy
import sys
import os
import socket
import gc
import time
import numpy as np
import logging

# Set overwrite option
arcpy.env.overwriteOutput = True
# datapath
DATA_DIR = r"C:\Data"
WORK_DIR = r"C:\Work01"
MODIS = DATA_DIR + "\\" + "MODIS_2001.gdb"
INDEX_POLY = DATA_DIR + "\\" + "MODIS_index.gdb"
def proc_zonalStat(feature_lyr, feature_path, list_index):
   arcpy.env.workspace = "in_memory"
   arcpy.SelectLayerByLocation_management(list_index[0], "INTERSECT", feature_lyr)
   raslist_modis = []
   cursor = arcpy.da.SearchCursor(list_index[0], ["F_NAME"])
   for row in cursor:
      raslist_modis.append(MODIS + "\\" + row[0])
      del cursor
ras_feature = "in_memory"+ "\\ras_feature"
arcpy.env.outputCoordinateSystem = arcpy.SpatialReference(4326)
arcpy.PolygonToRaster_conversion(feature_lyr, "OBJECTID", ras_feature, cellsize=0.0041666667)
time_1 = time.time() - start_time
logging.info("time_1:{0}".format(time_1) + "[sec]")
count = 0
aggreFlg = True
for inValueRaster in raslist_modis:
   try:
      tableName = "MODIS_tbl" + str(count).zfill(3)
      outCombine = arcpy.sa.Combine([ras_feature, inValueRaster])
      arcpy.TableToTable_conversion(outCombine, scratchGdb, tableName)
      arcpy.Delete_management(outCombine)
      count += 1
      time_2 = time.time() - start_time
      logging.info("time_2:{0}".format(time_2) + "[sec]")
fieldName = "MODIS_254"
expression = "0"
arcpy.CalculateField_management(feature_lyr, fieldName, expression, "PYTHON_9.3")
arcpy.Delete_management(ras_feature)
tables = arcpy.ListTables("MODIS_tbl*")
list_df_MODIS = []
fields = ['ras_feature', 'MODIS_landCover_', 'COUNT']
for table in tables:
   # numpy array -> pandas dataframe
   arr = arcpy.da.TableToNumPyArray(table, fields)
   df = pd.DataFrame(arr, columns=fields)
   list_df_MODIS.append(df)
   del arr
   gc.collect()
del tables
gc.collect()
df_all = pd.concat(list_df_MODIS, ignore_index=True)
add = pd.DataFrame([[0,0,0],
[0,1,0],
[0,2,0],
[0,3,0],
[0,4,0],
[0,5,0],
[0,6,0],
[0,7,0],
[0,8,0],
[0,9,0],
[0,10,0],
[0,11,0],
[0,12,0],
[0,13,0],
[0,14,0],
[0,15,0],
[0,16,0],
[0,254,0]],
columns=['ras_feature', 'MODIS_landCover_', 'COUNT'])
df_all = df_all.append(add)
del add
grouped = df_all.groupby([df_all['ras_feature'], df_all['MODIS_landCover_']], as_index=False).sum()
arr_MODIS = grouped.pivot_table(index='ras_feature', columns=['MODIS_landCover_'],values=['MODIS_landCover_','COUNT'],fill_value=0)
arr_MODIS = arr_MODIS.reset_index().values
arr_MODIS = arr_MODIS.astype(np.float32)
struct_array = np.core.records.fromarrays(arr_MODIS.transpose(), 
np.dtype([('Value', np.uint64), 
('MODIS_0', np.float32), 
('MODIS_1', np.float32), 
('MODIS_2', np.float32), 
('MODIS_3', np.float32), 
('MODIS_4', np.float32), 
('MODIS_5', np.float32), 
('MODIS_6', np.float32), 
('MODIS_7', np.float32), 
('MODIS_8', np.float32), 
('MODIS_9', np.float32), 
('MODIS_10', np.float32), 
('MODIS_11', np.float32), 
('MODIS_12', np.float32), 
('MODIS_13', np.float32), 
('MODIS_14', np.float32), 
('MODIS_15', np.float32), 
('MODIS_16', np.float32), 
('MODIS_254', np.float32)]))
tmp_table = tmp_proc_Gdb + "\\tmp_table"
arcpy.da.NumPyArrayToTable(struct_array, tmp_table)
del struct_array, grouped, arr_MODIS, df_all, list_df_MODIS
gc.collect()
tmp_table_view = tmp_proc_Gdb + "\\tmp_table_view"
arcpy.MakeTableView_management(tmp_table, tmp_table_view, "NOT Value = 0") 
arcpy.AddJoin_management(feature_lyr, "OBJECTID", tmp_table_view, "Value")
feature_name = os.path.basename(feature_path)
for i in range(0,17):
   fieldName = feature_name + ".MODIS_" + str(i)
   expression = "!tmp_table.MODIS_" + str(i) + "!"
   arcpy.CalculateField_management(feature_lyr, fieldName, expression, "PYTHON_9.3")

fieldName = feature_name + ".MODIS_254"
expression = "!tmp_table.MODIS_254!"
arcpy.CalculateField_management(feature_lyr, fieldName, expression, "PYTHON_9.3")
arcpy.RemoveJoin_management(feature_lyr)
arcpy.Delete_management(tmp_table_view)
arcpy.Delete_management(tmp_table)
time_3 = time.time() - start_time
logging.info("time_3:{0}".format(time_3) + "[sec]")


# Set scratchWorkspace
arcpy.env.scratchWorkspace = "in_memory"
# Set local_outGDB
THREAD = 0
hostName = socket.gethostname()
local_outGdb_name = 'tmp_' + hostName + '_' + str(THREAD).zfill(2) + '.gdb'
scratchGdb = "in_memory"
local_outGdb = WORK_DIR + '\\' + local_outGdb_name
tmp_proc_Gdb = "in_memory"

logfile_name = WORK_DIR + "\\log_test.txt"
logging.basicConfig(filename=logfile_name,level=logging.DEBUG)
logging.info("************** MAIN PROCESS START **************")
arcpy.env.workspace = WORK_DIR
for gdb in arcpy.ListFiles("test.gdb"):
if not arcpy.Exists(WORK_DIR + '\\' + local_outGdb_name):
   arcpy.CreateFileGDB_management(WORK_DIR, local_outGdb_name)

if arcpy.Exists(WORK_DIR + '\\Res\\' + gdb):
   arcpy.env.workspace = WORK_DIR + '\\Res\\' + gdb
   features_end = arcpy.ListFeatureClasses()

   arcpy.env.workspace = WORK_DIR + '\\' + gdb
   features_tmp = arcpy.ListFeatureClasses()

   features = [ feature for feature in features_tmp if feature not in features_end ]
else:
   arcpy.env.workspace = WORK_DIR + '\\' + gdb
   features = arcpy.ListFeatureClasses()
totalNum = len(features)
counter = 0

index_MODIS_lyr = "in_memory" + "\\" + "idx_MODIS_lyr"
list_index = [index_MODIS_lyr]
arcpy.MakeFeatureLayer_management(INDEX_POLY + "\\MODIS_landCover_2001", index_MODIS_lyr)
for feature in features:
   try:
      start_time = time.time()
      arcpy.env.workspace = WORK_DIR + '\\' + gdb
      arcpy.env.parallelProcessingFactor = "100%"
      counter += 1
      logging.info(feature+" "+str(counter) + '/' + str(totalNum)+" START")
      tmp_feature = tmp_proc_Gdb + "\\tmp_feature"
      arcpy.CopyFeatures_management(feature, tmp_feature)
      feature_lyr = tmp_proc_Gdb + "\\feature_lyr"
      arcpy.MakeFeatureLayer_management(tmp_feature, feature_lyr)
      rows = arcpy.GetCount_management(feature_lyr)
      countFeature = int(rows.getOutput(0))
      del rows

      arcpy.AddField_management(feature_lyr, "MODIS_0" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_1" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_2" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_3" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_4" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_5" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_6" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_7" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_8" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_9" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_10" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_11" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_12" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_13" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_14" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_15" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_16" , "FLOAT")
      arcpy.AddField_management(feature_lyr, "MODIS_254" , "FLOAT")
      proc_zonalStat(feature_lyr, tmp_feature, list_index)
      out_feature = local_outGdb + "\\" + feature
      arcpy.CopyFeatures_management(tmp_feature, out_feature)
      arcpy.Delete_management(feature_lyr)
      arcpy.Delete_management(tmp_feature)
      elapsed_time = time.time() - start_time
      logging.info("elapsed_time:{0}".format(elapsed_time) + "[sec]")
   except arcpy.ExecuteError:
      logging.info(arcpy.GetMessages(2))
      arcpy.Delete_management("in_memory")
      arcpy.MakeFeatureLayer_management(INDEX_POLY + "\\MODIS_landCover_2001", index_MODIS_lyr)
outDir = WORK_DIR + "\\Res"
outGdbPath = outDir + "\\" + gdb
if not arcpy.Exists(outGdbPath):
   arcpy.CreateFileGDB_management(outDir, gdb)
arcpy.env.workspace = local_outGdb
features = arcpy.ListFeatureClasses()
arcpy.FeatureClassToGeodatabase_conversion(features, outGdbPath)
arcpy.Delete_management(local_outGdb)
logging.info("************** MAIN PROCESS E N D **************")‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍

[YasunariMorita]

Additional information

I changed python version from 3.X to 2.7(changed Arc from pro to Map).

This slow down has not happened!

Is this the matter of ArcGIS pro 2.0???

[MarcoBoeringa]

I don't think this is necessarily a problem of Pro, nor of the Windows or Python / arcpy version. I have seen similar slowing down of equivalent looping processes in a huge model I developed with continued processing, but this issues occurs across the board, not depended on application, Windows or arcpy version.

The main thing you really want to do is to switch of the "Log geoprocessing operations to a log file" option. You can find it (in ArcMap), in the Geoprocessing / Geoprocessing / Results Management dialog.

As this Help page clearly states:

Viewing tool execution history—Help | ArcGIS for Desktop 

the logging can seriously affect tool performance with repeated iterator like looping.

Another aspect to be aware off is the tracking of the geoprocessing history and display of all the messages in the progress and results dialog of ArcMap / Pro. By default, I set the Geoprocessing / Geoprocessing / Results Management option in ArcMap to "Keep results younger than: Never save", because if I don't, and try to save my Map Document in ArcMap, it ends up being a bloated > 50 MB mxd file that takes two to three minutes or so to save. This is because the entire geoprocessing history is written in the map document.

[YasunariMorita]

Thank you for your information.

According to your advice, I added the code below.

arcpy.SetLogHistory(False)

SetLogHistory:

http://desktop.arcgis.com/ja/arcmap/10.3/analyze/arcpy-functions/setloghistory.htm

Then, you are right.Log setting caused mainly the issue.The result is

• first time:about10sec
• 51st time:about 25sec(improved!)

But,  the itrate process still slightly slow down...

It can not be helped,can it?

[MarcoBoeringa]

Then, you are right.Log setting caused mainly the issue.The result is

• first time:about10sec
• 51st time:about 25sec(improved!)

But,  the itrate process still slightly slow down...

It can not be helped,can it?

Well, yes and no. Obviously, if your experience with this particular script in Windows 7 was a different one, with almost no detectable slow down in both ArcMap and Pro, than at least theoretically, something potentially could be done you would assume...

But on the other hand, I do think there is a small inherent price to pay for the processing in ArcPy, the geoprocessing framework of ArcGIS, and ModelBuilder, where it is likely ArcGIS needs to keep track of stuff in order to do its thing, which will accumulate with an increased number of iterations in a script or Model.

[andysp]

Thanks so much for this info. I have been troubleshooting performance issues with TruncateTable, Append, AddField, and AddXY taking longer than seemed appropriate for the number of rows being processed. In one particular script, each was taking ~90 sec. With arcpy.SetLogHistory(False) in place, each statement takes <= 0.5 sec.

[DanPatterson_Retired]

I notice a lot of stuff use pandas.  Have you considered refactoring those sections just to use numpy.  Working with a structured array in numpy is can be faster than its pandas equivalent.  Also there is extra baggage in using a recarray rather than a structured array.  Recarrays do allow you to use object.property calls but numpy's object ['property'] method removes an extra layer of baggage as well.