The
main advantage of real-time data acquisition (RDA) is that new data is
reflected in your BI reports just a few minutes after being entered in
your operational systems. RDA therefore supports your business users to
make their tactical decisions on a day-by-day basis. The drawback
however is that these business users notice much faster when one of
their BI reports is not up to date. They might call you then and ask why
the document posted 5 minutes ago is not visible yet in reporting. And
what do you do now? I’ll show you how BW 7.3 helps you to resolve
problems with real-time data acquisition faster than ever before.
First,
let’s have a look at what else is new to RDA in BW 7.3. The most
powerful extension is definitely the HybridProvider. By using RDA to
transfer transactional data into a HybridProvider, you can easily
combine the low data latency of RDA with the fast response times of an
InfoCube or a BWA index, even for large amounts of data. You’ll find
more information about this combination in a separate blog.
Additionally. BW 7.3 allows for real-time master data
acquisition. This means that you can transfer delta records to
InfoObject attributes and texts at a frequency of one per minute. And
just like RDA directly activates data transferred to a DataStore object,
master data transferred to an InfoObject becomes available for BI
reporting immediately.
But
now, let’s start the RDA monitor and look at my examples for RDA
troubleshooting. I’ve chosen some data flows from my BW 7.0 test content
and added a HybridProvider and an InfoObject. I know that this flight
booking stuff is not really exciting, but the good thing is that I can
break it without getting calls from business users.
Remember
that you can double-click on the objects in the first column to view
details. You can look up for example that I’ve configured to stop RDA
requests after 13 errors.
Everything
looks fine. So let’s start the RDA daemon. It will execute all the
InfoPackages and DTPs assigned to it at a frequency of one per minute.
But wait… what’s this?
The
system asks me whether I’d like to start a repair process chain to
transfer missing requests to one of the data targets. Why? Ah, okay…
I’ve added a DTP for the newly created HybridProvider but forgotten to
transfer the requests already loaded from the DataSource. Let’s have a
closer look at these repair process chains while they are taking care of
the missing requests.
On
the left hand side, you can see the repair process chain for my
HybridProvider. Besides the DTP, it also contains a process to activate
DataStore object data and a subchain generated by my HybridProvider to
transfer data into the InfoCube part. On the right hand side, you can
see the repair process chain for my airline attributes which contains an
attribute change run. Fortunately, you don’t need to bother with these
details – the system is doing that for you. But now let’s really start
the RDA daemon.
Green
traffic lights appear in front of the InfoPackages and DTPs. I refresh
the RDA monitor. Requests appear and show a yellow status while they
load new data package by package. The machine is running and I can go
and work on something else now.
About a day later, I start the RDA monitor again and get a shock. What has happened?
The
traffic lights in front of the InfoPackages and DTPs have turned red.
The RDA daemon is showing the flash symbol which means that is has
terminated. Don’t panic! It’s BW 7.3. The third column helps me to get a
quick overview: 42 errors have occurred under my daemon, 4 DTPs have
encountered serious problems (red LEDs), and 4 InfoPackages have
encountered tolerable errors (yellow LEDs). I double-click on “42” to
get more details.
Let’s
have a short break to draw a comparison. What would you do in BW 7.0?
1) You could double-click on a failed request to analyze it. This is
still the best option to analyze the red DTP requests in our example.
But you could not find the tolerable RFC problems and runtime errors.
2)
You could browse through the job overview and the job logs. This would
have been the preferable approach to investigate the runtime errors in
our example. The job durations and the timestamps in the job log also
provide a good basis to locate performance issues, for example in
transformations.
3)
You could browse through the application logs. These contain more
details than the job logs. The drawback however is that the application
log is lost if runtime errors occur.
These
three options are still available in BW 7.3 – they have even been
improved. In particular, the job and application logs have been reduced
to the essential messages. Locating a problem is still a cumbersome task
however if you don’t know when it occurred. The integrated error
overview in the RDA monitor, BW 7.3 allows you to analyze any problem
with the preferred tool. Let me show you some examples.
Unless you
have other priorities from your business users I’d suggest starting
with runtime errors because they affect all objects assigned to the
daemon. RDA background jobs are scheduled with a period of 15 minutes to
make them robust against runtime errors. In our example, this means the
RDA daemon serves all DataSources from the one with the lowest error
counter up to the one which causes the runtime. The job is then
terminated and restarted 15 minutes later. The actual frequency is thus
reduced from 60/h to 4/h, which is not real-time anymore. Let’s see what
we can do here. I’ll double-click on “10” in the error column for the
request where the problem has occurred.
Just double-click on the error message in the overview to analyze the short dump.
Puh…
This sounds like sabotage! How can I preserve the other process objects
assigned to the same daemon from this runtime error while I search for
the root cause? I could just wait another hour of course. This RDA
request will then probably have reached the limit of 13 errors that I
configured with the InfoPackage. Once this threshold is reached, the RDA
daemon will exclude this InfoPackage from execution. The smarter
alternative is to temporarily stop the upload and delete the assignment
to the daemon.
The
overall situation becomes less serious once the DataSource has been
isolated under “Unassigned Nodes”. The daemon continues at a frequency
of onc per minute although there are still 32 errors left.
Note
that most of these errors – namely the RFC failures – can be tolerated.
This means that these errors (yellow LEDs) do not hinder InfoPackages
or DTPs until the configured error limit is reached. Assume that I’ve
identified the root cause for the RFC failures as a temporary issue. I
should then reset the error counter for all objects that have not
encountered other problems. This function is available in the menu and
context menu. The error counter of an InfoPackage or DTP is reset
automatically when a new request is created. Now let’s look at one of
the serious problems. I’ll therefore double-click on “2” in the error
column of the first DTP with red LED.
When
I double-click on the error message, I see the exception stack
unpacked. Unfortunately that does not tell me more than I already knew:
An exception has occurred in a sub step of the DTP. So I navigate to the
DTP monitor by double-clicking the request ID (217).
Obviously,
one of the transformation rules contains a routine that has raised the
exception “13 is an unlucky number”. I navigate to the transformation
and identify the root cause quickly.
In
the same way, I investigate the exception which has occurred in DTP
request 219. The DTP monitor tells me that something is wrong with a
transferred fiscal period. A closer look at the transformation reveals a
bug in the rule for the fiscal year variant. Before I can fix the
broken rules, I need to remove the assignment of the DataSource to the
daemon. When the corrections are done, I schedule the repair process
chains to repeat the DTP requests with the fixed transformations.
Finally I re-assign the DataSource to the daemon.
The RDA monitor
already looks much greener now. Only one DataSource with errors is
left. More precisely, there are two DTPs assigned to this DataSource
which encountered intolerable errors, so the request status is red.
Again, I double-click in the error column to view details.
The
error message tells me straight away that the update command has caused
the problem this time rather than the transformation. Again, the DTP
monitor provides insight into the problem.
Of
course “GCS” is not a valid currency (Should that be “Galactic Credit
Standard” or what?). I go back to the RDA monitor and double-click on
the PSA of the DataSource in the second column. In the request overview,
I mark the source request of the failed DTP request and view the
content of the problematic data package number 000006.
Obviously,
the data is already wrong in the DataSource. How could this happen? Ah,
okay… It’s an InfoPackage for Web Service (Push). Probably the source
is not an SAP system, and a data cleansing step is needed – either in
the source system or in the transformation. As a short-term solution, I
could delete or modify the inconsistent records and repeat the failed
DTP requests with the repair process chain.
That’s all. I hope
you enjoyed this little trip to troubleshooting real-time data
acquisition, even though this is probably not part of your daily work
yet. Let me summarize what to do if problems occur with RDA. Don’t
panic. BW 7.3 helps you to identify and resolve problems faster than
ever before. Check the error column in the RDA monitor to get a quick
overview. Double-click wherever you are to get more details. Use the
repair process chains to repeat broken DTP requests.
No comments:
Post a Comment