This is a set of scripts for collecting and processing Oracle ASM Metrics.
When the data is collected from the ASM instance, data for all disks, diskgroups and databases is collected.
As this is file based CSV data, it is fairly easy to filter and aggregate the data as needed.
Following are descriptions of many of the scripts. A later section has examples of processing the data and charting it with Excel.
Note: These are Bash scripts, and as you might expect, these work on Linux and other Unix like OS's.
The Excel charting requires Perl with the Excel::Writer::XLSX module installed.
Of course, you can always just manually load the CSV into Excel for charting.
Collect ASM metrics for further processing.
Use asm-metrics-collector.sh
bash driver example
Calls all of the following scripts in the correct order.
This script aggregates metrics based on one column from raw metrics data. eg. Aggregate IO metrics per diskgroup
bash driver example - aggregates by diskgroup
bash driver example - aggregates by disk
Create Excel file with charts of metrics
bash driver example using aggregates by diskgroup
bash driver example using aggregates by diskname
Perl script used to look for negative values in raw CSV files. Used for some debugging. This script can be used to look for any value with minor adjustment
Create a histogram of ASM read or write time metrics.
asm-time-histogram.sh -h
Calls asm-time-histogram.sh to create histograms of reads/writes for all disk groups
Calls asm-time-histogram.sh to create histograms of reads/writes for all disks per each disk group
This script will normalize disk read or write times into buckets of 5ms and count them.
The output can be sent to dynachart.pl for charting.
This script calls disk-time-histogram.pl for a series of diskgroups and read/write times.
The output is sent to dynachart.pl and Excel spreadsheets with scatter plots created.
The output is also suitable for histograms.
This script reads STDIN, expecting a timestamp, disk number and read or write tim.
Rows are then arranged into columns:
timestamp,disk 1, disk 2,...
2022-03-09 15:00:00, 0.001, 0.025,...
ie.
Get disktimes per disk for instance 1, the DATA diskgroup:
$ grep -ahE 'orcl01,orcl,.*,DATA,' <(tail -q -n+2 logs/asm-*.csv) | cut -d, -f1,7,12 | ./disk-group-by-timestamp.pl > db-disk-breakout/db007-data001-readtim.csv
This script will generate a sed file `anonymize.sed' containing the names of diskgroups.
As generated, nothing will actually be changed.
Edit the file to change the names as you would like.
Here is an example file that was just generated:
s/\b_mgmtdb/_mgmtdb/g
s/\MYDB/MYDB/g
s/\bMYCORP_DATA/MYCORP_DATA/g
s/\bFRA_MYCORP_01/FRA_MYCORP_01/g
s/\bDATA_MYCORP_01/DATA_MYCORP_01/g
s/\bREDO_MYCORP_01/REDO_MYCORP_01/g
s/\bREDO_MYCORP_02/REDO_MYCORP_02/g
After editing:
s/\b_mgmtdb/_mgmtdb/g
s/\MYDB/ORCL/g
s/\bMYCORP_DATA/MDATA/g
s/\bFRA_MYCORP_01/FRA/g
s/\bDATA_MYCORP_01/DATA/g
s/\bREDO_MYCORP_01/REDO_01/g
s/\bREDO_MYCORP_02/REDO_02/g
Then just run the sed script for each CSV file.
The CSV files will not be backed up, so don't use this on the only copy of the CSV files.
for f in *.csv
do
echo "working on $f"
sed -i -r -f anonymize.sed $f
done
This script calculates avg read/write times per diskgroup.
Adjust values in the script at as needed
bash script used for verifying aggregates during development
asm-metrics-collector.pl had a bug where HH was used in the timestamp format when really it should have been HH24.
This bug resulted in PM timestamps that were 12 hours off. In addition when the time changed from 12:59 to 13:00, this bug resulted in negative elapsed times.
This script will fix the snaptime and elapsed time in output files. A new filename with '-corrected' inserting in the name will be created.
example:
asm-metrics-fix.pl asm-metrics-1.csv asm-metrics-2.csv asm-metrics-3.csv
Three new corrected files will be created: asm-metrics-1-corrected.csv asm-metrics-2-corrected.csv asm-metrics-3-corrected.csv
After running asm-metrics-aggregator_bydisk.sh, use reads-per-disk.sh to create a log file with a histogram of reads for each disk in the diskgroup.
After running asm-metrics-aggregator_bydisk.sh, use writes-per-disk.sh to create a log file with a histogram of writes for each disk in the diskgroup.
Data is raw. Metrics are per diskgroup and disk every 20 seconds.
So, one line for each disk.
For analysis and charting, the data is being aggregated.
Some detail is lost, but this level of detail is not necessary for charting trends.
Make a copy of asm-metrics-aggregator.sh, and modifify the default file names.
Or to run against an entire directory of files , use asm-metrics-aggregator-loop.sh instead.
To see the effect, the asm-metrics-aggregator.sh script was used to process 10000 lines of ASM metrics data in the file before.csv.
wc -l before.csv after.csv
10000 before.csv
51 after.csv
10051 total
The metrics are provided per diskgroup, per snapshot time.
The metric values have all been summed, so the the average values can be calculated as needed.
$ head -5 before.csv
DISPLAYTIME,SNAPTIME,ELAPSEDTIME,INSTNAME,DBNAME,GROUP_NUMBER,DISK_NUMBER,DISKGROUP_NAME,READS,WRITES,READ_TIME,AVG_READ_TIME,WRITE_TIME,AVG_WRITE_TIME,BYTES_READ,BYTES_WRITTEN,COLD_BYTES_READ,COLD_BYTES_WRITTEN,COLD_READS,COLD_USED_MB,COLD_WRITES,CREATE_DATE,DISK_NAME,FAILGROUP,FAILGROUP_TYPE,FREE_MB,HEADER_STATUS,HOT_BYTES_READ,HOT_BYTES_WRITTEN,HOT_READS,HOT_USED_MB,HOT_WRITES,LABEL,MOUNT_DATE,OS_MB,PATH,PREFERRED_READ,PRODUCT,READ_ERRS,REDUNDANCY,REPAIR_TIMER,SECTOR_SIZE,TOTAL_MB,UDID,VOTING_FILE,WRITE_ERRS
2020-07-31 08:55:35,2020-07-31 08:55:35.629847,20.444252,PRDBTS13,PRDBTS1,3,24,DATA,0,206,0,0,0.483442000004288,0.00234680582526354,0,1843200,0,1843200,0,1673632,206,2018-04-13 13:46:15,DATA_0024,DATA_0024,REGULAR,374540,MEMBER,0,0,0,48980,0,,2019-10-12 22:52:43,2097152,/dev/mapper/asm_data_25p1,U,,0,UNKNOWN,0,512,2097152,,N,0
2020-07-31 08:55:35,2020-07-31 08:55:35.629847,20.444252,PRDBTS11,PRDBTS1,3,48,DATA,0,4,0,0,0.003153999998176,0.000788499999544001,0,49152,0,49152,0,1673236,4,2020-02-25 15:53:42,DATA_0048,DATA_0048,REGULAR,374916,MEMBER,0,0,0,49000,0,,2020-02-25 15:53:44,2097152,/dev/mapper/asm_data_49p1,U,,0,UNKNOWN,0,512,2097152,,N,0
2020-07-31 08:55:35,2020-07-31 08:55:35.629847,20.444252,PRDBTS14,PRDBTS1,4,0,FRA,0,0,0,0,0,0,0,0,0,0,0,504391,0,2017-06-14 13:54:53,FRA_0000,FRA_0000,REGULAR,1592761,MEMBER,0,0,0,0,0,,2019-10-12 22:52:43,2097152,/dev/mapper/asm_fra_1p1,U,,0,UNKNOWN,0,512,2097152,,N,0
2020-07-31 08:55:35,2020-07-31 08:55:35.629847,20.444252,PRDBTS14,PRDBTS1,6,7,REDO,752,8,0.343787999998312,0.000457164893614777,0.0216940000009345,0.00271175000011681,12435456,131072,12435456,131072,752,55,8,2020-02-26 11:38:29,REDO_0007,REDO_0007,REGULAR,20865,MEMBER,0,0,0,81480,0,,2020-02-26 11:38:29,102400,/dev/mapper/asm_redo_8p1,U,,0,UNKNOWN,0,512,102400,,N,0
$ head -6 after.csv
DISPLAYTIME,ELAPSEDTIME,DISKGROUP_NAME,READS,WRITES,READ_TIME,WRITE_TIME,BYTES_READ,BYTES_WRITTEN,READ_ERRS,WRITE_ERRS
2020-07-31 08:55:35,20.444252,FRA,80,807,31.6627970000747,46.6677670000765,1671680,21086720,0,0
2020-07-31 08:55:35,20.444252,GRID,64,196,0.0423570000002656,0.273379000002024,991232,8220672,0,0
2020-07-31 08:55:35,20.444252,ACFS,0,0,0,0,0,0,0,0
2020-07-31 08:55:35,20.444252,REDO,6259,43555,781.33763800023,64.7069339999862,436242432,336217088,0,0
2020-07-31 08:55:35,20.444252,DATA,187799,88229,1004.5424800001,875.308896999404,47032052224,1512292352,0,0
Usually asm-metrics-aggregator-loop.sh script is used, which reads files from the logs directory, and writes to the output directory.
So as to avoid doing this step in Excel, Google Sheets, or other charting tools, the averages can be calcualted for average read and write times, and written to a new file.
Some of the scripts that do this:
- asm-metrics-synth.pl
- asm-metrics-synth.sh
The asm-metrics-synth.sh script is just a driver for asm-metrics-synth.pl, and gets its input from the output directory of the previous step.
New files are written to the synth directory.
Let's try a test on the after.csv file created earlier
$ ./asm-metrics-synth.pl after.csv synth.csv
> head -6 after.csv synth.csv
==> after.csv <==
DISPLAYTIME,ELAPSEDTIME,DISKGROUP_NAME,READS,WRITES,READ_TIME,WRITE_TIME,BYTES_READ,BYTES_WRITTEN,READ_ERRS,WRITE_ERRS
2020-07-31 08:55:35,20.444252,FRA,80,807,31.6627970000747,46.6677670000765,1671680,21086720,0,0
2020-07-31 08:55:35,20.444252,GRID,64,196,0.0423570000002656,0.273379000002024,991232,8220672,0,0
2020-07-31 08:55:35,20.444252,ACFS,0,0,0,0,0,0,0,0
2020-07-31 08:55:35,20.444252,REDO,6259,43555,781.33763800023,64.7069339999862,436242432,336217088,0,0
2020-07-31 08:55:35,20.444252,DATA,187799,88229,1004.5424800001,875.308896999404,47032052224,1512292352,0,0
==> synth.csv <==
DATE,TIME,ELAPSEDTIME,DISKGROUP_NAME,AVG_READ_TIME,AVG_WRITE_TIME,READS,WRITES,READ_TIME,WRITE_TIME,BYTES_READ,BYTES_WRITTEN,READ_ERRS,WRITE_ERRS
2020-07-31,08:55:35,20.444252,FRA,0.395785,0.057829,80,807,31.6627970000747,46.6677670000765,1671680,21086720,0,0
2020-07-31,08:55:35,20.444252,GRID,0.000662,0.001395,64,196,0.0423570000002656,0.273379000002024,991232,8220672,0,0
2020-07-31,08:55:35,20.444252,ACFS,0,0,0,0,0,0,0,0,0,0
2020-07-31,08:55:35,20.444252,REDO,0.124834,0.001486,6259,43555,781.33763800023,64.7069339999862,436242432,336217088,0,0
2020-07-31,08:55:35,20.444252,DATA,0.005349,0.009921,187799,88229,1004.5424800001,875.308896999404,47032052224,1512292352,0,0
There are 2 changes made to the synth.csv file.
- DISPLAYTIME has been split into DATE and TIME columns
- New columns are AVG_READ_TIME and AVG_WRITE_TIME
To further simplify charting, the data can be split into diskgroup based files.
The asm-diskgroup-breakout.sh is a fairly simple Bash script that creates diskgroup based files, taking its input from the synth directory of the previous step.
New files are written to the diskgroup-breakout directory.
Sometimes there may be outliers and very large spikes. These tend to skew the data and make the charts hard to read.
In extreme cases a chart may consist of some spike connected by a flat line.
Normally you would like to see that data charted.
The data for each file is piped through two Python scripts, outlier-remove.py and flatten.py.
These CSV files and XLS files are placed in directories with a suffix of '-chart'.
Should you not want these, simply comment out the asm-*clean*.sh lines in run-std-asm.sh.
Using the disk-pivot.pl script, transform this layout:
timestamp diskname metric1 metric2 metric3
To this layout:
timestamp disk1 disk2 disk3 ... metric
The default metric is avg_wrt_time, but it can be any metric, just modify the code a bit as needed.
The data can now be charted using asm-metrics-chart.pl.
There are already a couple of shell driver scripts for this task:
- asm-metrics-chart.sh
- asm-metrics-chart-synth.sh
This script is hardcoded to create a single Excel file from an ASM Metrics file.
It can be used as the basis for a script that processes an entire directory of files.
This script is configured to process all the csv files in a single directory.
The input is taken from the diskgroup-breakout directory created in the previous step.
The output is Excel files, written to the xlsx-by-diskgroup directory.
The commands to process all data were put into a the Bash script, run-std-asm.sh.
#!/usr/bin/env bash
./asm-metrics-aggregator-loop.sh
./asm-metrics-synth.sh
./asm-diskgroup-breakout.sh
./asm-metrics-chart-synth.sh$ time ./run-std-asm.sh
working on output/asm-data-20200731-085514.csv
working on output/asm-data-20200801-084753.csv
working on output/asm-data-20200802-083738.csv
working on output/asm-data-20200803-083430.csv
working on output/asm-data-20200804-083016.csv
working on output/asm-data-20200805-083703.csv
working on synth/asm-data-20200731-085514.csv
working on synth/asm-data-20200801-084753.csv
working on synth/asm-data-20200802-083738.csv
working on synth/asm-data-20200803-083430.csv
working on synth/asm-data-20200804-083016.csv
working on synth/asm-data-20200805-083703.csv
working on synth/asm-data-20200731-085514.csv
working on synth/asm-data-20200801-084753.csv
working on synth/asm-data-20200802-083738.csv
working on synth/asm-data-20200803-083430.csv
working on synth/asm-data-20200804-083016.csv
working on synth/asm-data-20200805-083703.csv
working on xlsx-by-diskgroup/ACFS.xlsx
working on xlsx-by-diskgroup/DATA.xlsx
working on xlsx-by-diskgroup/FRA.xlsx
working on xlsx-by-diskgroup/GRID.xlsx
working on xlsx-by-diskgroup/REDO.xlsx
real 4m53.885s
user 3m44.880s
sys 0m23.900s
ASM Metrics files are to be organized like this:
cluster-01/
node-01/logs
node-02/logs
node-03/logs
cluster-02
node-01/logs
node-02/logs
node-03/logs
node-04/logs
...
Then:
./runall-cluster.sh
The data is collected at the ASM device level.
This means that data can be sliced and diced many ways.
For instance, take a look at the dg-diskgroup scripts:
$ ls -1 *db-diskgroup*
asm-db-diskgroup-breakout.sh
asm-metrics-aggregator-db-diskgroup.sh
asm-metrics-chart-db-diskgroup-synth.sh
asm-metrics-chart-db-diskgroup.sh
asm-metrics-db-diskgroup-synth.pl
asm-metrics-db-diskgroup-synth.sh
run-db-diskgroup.sh
These scripts can be used to create charts for each database and diskgroup pairing.
For instance, if there are 42 databases sharing the +DATA diskgroup, then 42 differenct XLXS files will be created, each showing the individual contribution toward all IO seen in +DATA.
Various utility scripts
Used to derive IOPS info from the diskgroup-breakout CSV files
Called by asm-metrics-add-iops.sh
Calls asm-metrics-add-iops.pl
Called in run-std-asm.sh
Calls dynachart.pl to create Excel files from the diskgroup-breakout-iops directory.
Scatter charts are created per CSV file, with the IOPS column on a secondary axis
Called by get-iops-disk-breakout.sh
Use this script to get IOPS from CSV files in the disk-breakout directories.
See get-iops.sh for a similar example.
Called by get-iops.sh
This script can be used to get IOPS values from raw asm-metrics CSV files.
$ ./get-iops.sh
cutFields: 1,3,8,9,10,15,16
###################################
## data set: set-01
###################################
dg PLXPRD01_DATA01 | reads: 301700269 | writes: 97035661 | time: 848760 | iops: 469
dg PLXPRD01_FRA01 | reads: 13962 | writes: 3098266 | time: 848760 | iops: 3
dg PLXPRD01_REDO01 | reads: 63116190 | writes: 101632161 | time: 848760 | iops: 194
dg PLXPRD01_REDO02 | reads: 315466 | writes: 101632172 | time: 848760 | iops: 120
dg PLXPRD02_DATA01 | reads: 280786418 | writes: 90006359 | time: 784213 | iops: 472
dg PLXPRD02_FRA01 | reads: 14458 | writes: 3020210 | time: 784213 | iops: 3
dg PLXPRD02_REDO01 | reads: 156300246 | writes: 101207900 | time: 784213 | iops: 328
dg PLXPRD02_REDO02 | reads: 285921 | writes: 101207912 | time: 784213 | iops: 129
dg PLXPRD03_DATA01 | reads: 121709752 | writes: 95920821 | time: 843191 | iops: 258
dg PLXPRD03_FRA01 | reads: 16032 | writes: 3263285 | time: 843191 | iops: 3
dg PLXPRD03_REDO01 | reads: 31235407 | writes: 108167077 | time: 843191 | iops: 165
dg PLXPRD03_REDO02 | reads: 215444 | writes: 108167121 | time: 843191 | iops: 128
###################################
## data set: set-03
###################################
dg ACFS | reads: 0 | writes: 0 | time: 861641 | iops: 0
dg DATA | reads: 8122420470 | writes: 2491225862 | time: 861641 | iops: 12317
dg FRA | reads: 90602141 | writes: 75719696 | time: 861641 | iops: 193
dg GRID | reads: 10783051 | writes: 11018609 | time: 861641 | iops: 25
dg REDO | reads: 358005076 | writes: 848432182 | time: 861641 | iops: 1400
This script came about as a result of investigating slow writes on a diskgroup.
Sometimes the avg write time would be 3 seconds, and only for slow writes.
This script can be used to show those writes, and the preceding 5 lines of data for that disk.
$ head -1 asm-data-20220110-065458.csv > asm-fra01-slow-writes.csv
$ grep -h ',dbp004,.*,FRA01,' asm-data*.csv | ./get-slow-writes-hist.pl >> asm-fra01-slow-writes.csv
Sample the data
$ cut -d, -f1,7,8,14 asm-fra01-slow-writes.csv | head -30
DISPLAYTIME,DISK_NUMBER,DISKGROUP_NAME,AVG_WRITE_TIME
2022-01-09 07:47:46,1,FRA01,0.151036499999464
2022-01-09 07:48:45,1,FRA01,0.00115562500104716
2022-01-09 07:49:44,1,FRA01,0.00105226829274637
2022-01-09 07:50:43,1,FRA01,0.00102099999882436
2022-01-09 07:51:42,1,FRA01,3.00137900000846
2022-01-09 07:50:43,5,FRA01,0.00106599999708124
2022-01-09 07:51:42,5,FRA01,0
2022-01-09 07:52:41,5,FRA01,0.429843142857343
2022-01-09 07:53:40,5,FRA01,0.0010293465345985
2022-01-09 07:54:39,5,FRA01,0.354050882353262
2022-01-09 07:55:38,5,FRA01,3.00099899999623
2022-01-09 07:50:43,6,FRA01,0.00122180000180379
2022-01-09 07:51:42,6,FRA01,0
2022-01-09 07:52:41,6,FRA01,0
2022-01-09 07:53:40,6,FRA01,0.00101174999984393
2022-01-09 07:54:39,6,FRA01,0.143833952380754
2022-01-09 07:55:38,6,FRA01,3.00096099999791
2022-01-09 07:50:43,7,FRA01,1.50122099999862
2022-01-09 07:51:42,7,FRA01,0
2022-01-09 07:52:41,7,FRA01,0.00101109090874988
2022-01-09 07:53:40,7,FRA01,0.00100167692309277
2022-01-09 07:54:39,7,FRA01,0.00100507258062979
2022-01-09 07:55:38,7,FRA01,3.00107300000673
2022-01-09 07:51:42,1,FRA01,3.00137900000846
2022-01-09 07:52:41,1,FRA01,0.0011576666632512
2022-01-09 07:53:40,1,FRA01,0.000990931404072498
2022-01-09 07:54:39,1,FRA01,0.023830571065986
2022-01-09 07:55:38,1,FRA01,0.00108950000139885
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent