@rob-clements
Forum Replies Created
-
I found this which may be helpful to others:
Run in a sub-thread option- When this option is selected on the Inbound TPS Editor by clicking Edit for TPS Inbound Data, a sub-thread is created under the protocol thread. This sub-thread is used only for data type messages, not those of type reply. There must be a TPS defined; there is no sub-thread if a TPS is not defined.
- There is only one thread involved with any IB/OB thread SMS (that is, TPS).
- All protocol threads have a pre-SMS queue and a post-SMS queue. If there is a TPS defined, messages go in the pre-SMS queue. The pre-SMS queue is skipped if no TPS is defined for the protocol thread. Between the two queues is the TPS that does something with the message and generates more messages.
- If a sub-thread is used, that TPS runs in it. During the time of the TPS processing, the engine is free to do other work. For example, input messages, output messages, move messages as part of the data flow. When the sub-thread has run the TPS procedure, it then signals the engine to pass the results into the post-SMS queue.
- If the defined TPS is small and fast, there is no requirement to use a sub-thread. The overhead of starting the thread would negate any gain. By “fast,” it is meant that it does not parse a message, as that can take time. It also does not access external data.
- In terms of stages, the IB pre and post queues are stages 1 and 2 and the OB queues are stages 10 and 11.
- In general, a single CPU machine gains slightly or even loses by turning on the sub-thread. This is the same as xlates.
- In the same way, a process that is lightly loaded does not gain anything using the sub-thread. There is no work for the engine to do when it runs.
- A sub-thread is helpful when there is a CPU available. It is also helpful when there is other work for the engine to do when the sub-thread is running. Parallel processing of messages is permitted if that is possible in the configuration.
- Note: When using Java TPS UPoCs, the Inbound tab shows the procedure as cljTPS. This is the Tcl interface procedure to Java TPS. When you click Edit, the Java class is shown. Java UPoCs cannot be used if Run in a Sub-thread is enabled
I also am having difficulty with the idea that even with the hcidbdump of our recovery and error databases for a given site (not the one with the huge files) messages are accumulating in the databases without reduction. I have an rLogM2K file on our site that handles document embedded messages (also, not the ones that are causing problems) and that’s 66mb and its been running for years using the standard cleanup process. I just cant see how the recovery database would be that relatively small if the only init we use is hcidbinit -iC. I think im missing a piece of this puzzle somewhere.
We absolutely do that. What we didn’t know is that the upstream source had started sending huge pdfs which went along the wire to the downstream receiver. The huge file would break the downstream receiver but they had an automated system the would “fix” the problem on their end after a certain amount of resends. So basically the interface would queue up and then dequeue all under our threshold of alerts. That’s the best we can figure out at least. We’ve been doing this for 20 years since quovadx and never had this type of problem before. We’ve really dug into the Db structure and are trying to understand if its better to fully clear all our databases, at least on the site with such large files coming through. Which is how I’m leaning now.
When I upgraded and migrated about 6 months ago I posed the HCILICSET question to support. Here is what they responded with:
Simply dropping the emailed license.dat file in to the vers folder is now the preferred method of applying a license.
I have this escalated to development. They have confirmed this is a known issue and have an AR# 17679 created for tracking. Regarding misleading instructions in documentation that is going to be corrected.
AR 21603 Remove hcilicset from install guide.
Even though this question is answered above, the bug information might be useful. And since it says use HCILICSET in the docs – those are officially wrong docs.
