mujiannan/Analysis-Services
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
rahulten 2020-11-12 17:19:46 -08:00 committed by GitHub
parent d566ad8e3b
commit 44fd16189c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 15 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
ASJobGraphEvents/README.md
View File

@ -1,10 +1,8 @@
# Job Graph Events in Power BI / Azure Analysis Services
Job Graph events can be used to identify bottlenecks in data refreshes by highlighting the critical path. For instances of Analysis Services not running on-premise, the graph is broken into 16 Kb chunks, each in their own event. The events can be reassembled with this script.
# Summary # Summary
The Job Graph event is emitted by AS Engine before and after processing. Both graphs show the order that jobs execute within the sequence point algorithm. The final (annotated) graph shows execution times for each job. In addition, the annotated graph shows the critical path. The Job Graph events are emitted by AS Engine before and after any processing (aka refresh) commands. Both job graphs show the order in which processing jobs execute within the sequence point algorithm, in the Analysis Services engine. The graph associated with the GraphEnd (2) Subclass is annotated, in that it contains the "Blocked duration", "Waiting duration", "Running duration" etc for each job. In addition, the annotated graph shows the critical path. Job Graph events can be used to identify bottlenecks in processing jobs, by highlighting the critical path.
In the cloud instances of Analysis Services, such as Azure Analysis Services or Power BI, the job graph is broken down into smaller chunks, each in their own event. The events can be reassembled with this script.
## Critical Path ## Critical Path
@ -22,25 +20,30 @@ The critical path is shown via dark backgrounds and white text in the job graph
## Usage ## Usage
1. Start a trace in SQL Server Profiler and select "Job Graph Events". 1. Start a trace in SQL Server Profiler and select "Job Graph Events".
2. Start a data refresh ("Process Full" in SQL Server Management Studio). 2. Initate processing (eg. "Process Full" in SQL Server Management Studio).
3. Wait for all trace events to arrive in Profiler. 3. Wait for the processing to finish
4. `File > Save As > Trace XML File` 4. Once the processing is finished, wait for all job graph events to show up in SQL Profiler. NOTE : The first event in a subclass is always the metadata, which contains, among other things, the number of additional events that'll follow.
5. Aim `rebuild.py` at this file like so: 5. Save all the events in a trace file, `File > Save As > Trace XML File`
6. Rebuild the annotated job graph by aiming the script `rebuild.py` at this file like so:
```bash ```bash
python rebuild.py path\to\trace.xml output_folder python rebuild.py path\to\trace.xml output_folder
``` ```
6. Inside `output_folder` there will be two .DGML files, which can be opened in Visual Studio. 7. Inside `output_folder` there will be two .DGML files, which can be opened in Visual Studio.
# Diagnosing from DGML # Diagnosing from DGML
Now that the graph is open, look for the critical path (darker nodes) or for failed jobs (white with a dashed border). Failed jobs include an "Error Code" that specifies why the job failed. Once the Job Graph is rendered, look for the critical path (darker nodes) or for failed jobs (white with a dashed border). Failed jobs include an "Error Code" that specifies why the job failed.
To diagnose slow refresh times, look for the critical path and start at the top. Look at "Blocked duration", "Waiting duration", and "Running duration". If a job has a long blocked duration, it spent a long time waiting on other jobs. If a job has a long waiting duration, it was waiting for an available thread, so increasing maxParallelism could help. If a job has a long running duration, then the model might have to be changed to speed up the job. To diagnose slow refresh times, look for the critical path and start at the top. Look at "Blocked duration", "Waiting duration", and "Running duration". If a job has a long blocked duration, it spent a long time waiting on other jobs. If a job has a long waiting duration, it was waiting for an available thread, so increasing maxParallelism could help. If a job has a long running duration, then the model might have to be changed to speed up the job.
# Creating a Gantt Chart # Creating a Gantt Chart
The Gantt Chart can also come in handy when diagnosing processing issues. Power BI Desktop provides first party and 3rd party Gantt Chart visuals, that can be used to visualize the job graph.
We have also provided a script to generate a Gantt Chart as a simple HTML file. Please follow the below instructions to generate the HTML
## Requirements ## Requirements
* Python 3.8 or later * Python 3.8 or later
@ -48,7 +51,7 @@ To diagnose slow refresh times, look for the critical path and start at the top.
## Usage ## Usage
1. Get a .DGML file with all the anntoations (running duration, waiting duration, etc.) 1. Get the reconstructed DGML file corresponding to the annotated graph, i.e the GraphEnd (2) Subclass
2. Run `gantt\script.py` like so: 2. Run `gantt\script.py` like so:
```bash ```bash