invocations | Reusable Invoke tasks

Like you noticed already, you can only check interactions on a mocked object type, i.e. mock, stub or spy. The latter, a spy, is what you need in this case, i.e. something like:

You must look at your usage profile. Do the request come constantly at a steady rate? Or are they spread out?

With a virtual machine you pay for the time it is running, it is not dependent on what it is doing.

With an Azure function consumption plan you pay per request. So when there are no requests there is no charge.

https://azure.microsoft.com/en-us/pricing/details/functions/ (Your 18 USD comes from this page?)

When a function has 500 MB to use, your code can use all the memory. When a VM has 500 MB of RAM a significant portion is used by the operating system.

Edit: As Ken mentioned in the comment with a VM you need to look after the server, so you also need to take that cost into consideration.

The compute capacity is the same given a steady constant continuous use where you turn off the VM when the 3M calls are finished. But the VM has additional costs that also need to be taken into consideration.

Note when you turn the VM off you still pay for the storage of the disks.

An explosion of errors and misguided ideas inherent in this question:

available() doesn't work. Well, it does what the javadoc says it does, and if you read the javadoc, and read it very carefully, you should come to the correct conclusion that what that is, is utterly useless. If you ever call available(), you've messed up. You're doing so here. More generally your read and write code doesn't work. For example, .read(byteArr) also doesn't do what you think it does. See below.

You can't 'save the state' of arbitrary objects, and if you want to push the idea, then if you can, then certainly not in the way you're doing it, and in general this is advanced java that involves hacking the JDK itself to get at it: Think of an InputStream that represents data flowing over a network connection. What do you imagine the 'serialization' of this InputStream object should look like? If you consider serialization as 'just represent the underlying data in memory', then what you'd get is a number that represents the OS 'pipe handle', and possibly some IP, port, and sequence numbers. This is a tiny amount of data, and all this data is completely useless - it doesn't say anything meaningful about that connection and this data cannot be used to reconstitute it, at all. Even within the 'scope' of a single session (i.e. where you serialize, and then deserialize almost immediately afterwards), as networks are a stream and once you grab a byte (or send a byte), it's gone. The only useful, especially for the notion of 'lets replay everything that happened as a test', serialization strategy involves actually 'recording' all the bytes that were picked up, as it happens, on the fly. This is not a thing that you can do as a 'moment in time' concept, it's continuous. You need a system that is recording all the things (it needs to be recording every inputstream, every outputstream, every time System.currentTimeMillis() in invoked, every time a random number is generated, etc), and then needs to use the results of recording it all when your API is asked to 'save' an arbitrary state.

Serialization instead is a thing that objects need to opt into, and where they may have to write custom code to properly deal with it. Not all objects can even be serialized (an InputStream representing a network pipe, as above, is one example of an object that cannot be serialized), and for some, serializing them requires some fancy footwork, and the only hope you have is that the authors of the code that powers this object put in that effort. If they didn't, there is nothing you can do.

The serialization framework of java awkwardly captures both of these notions. It does mean that your code, even if you fix the bugs in it, will fail on most objects that can exist in a JVM. Your testing tool can only be used to test the most simplistic code.

If you're okay with that, read on. But if not, you need to completely rethink what you're going to do with this.

This is not just my opinion, the openjdk team itself is broadly in agreement (they probably wouldn't quite put it like that, of course). The data emitted by OOS is a weird, inefficient, and underspecced binary blob. You can't analyse this data in any feasible way other than spending a few years reverse engineering the protocol, or just deserializing it (which requires having all the classes, and a JVM - this can be an acceptable burden, depends on your use case).

Contrast to e.g. Jackson which serializes data into JSON, which you can parse with your eyeballs, or in any language, and even without the relevant class files. You can construct 'serialized JSON' yourself without the benefit of first having an object (for testing purposes this sounds like a good idea, no? You need to test this testing framework too!).

If you understand all the caveats above and somehow still conclude that this project, as written and continuing to use the ObjectOutputStream API is still what you want to do (I really, really doubt that's the right call):

Use the newer APIs. available() does not return the size of that blob. read(someByteArray) is not guaranteed to fill the entire byte array. Just read the javadoc, it spells it out.

There is no way to determine the size of an inputstream by asking that inputstream. You may be able to ask the DB itself (usually, LENGTH(theBlobColumn) works great in a SELECT query.

If you somehow (e.g. using LENGTH(tbc)) know the full size, you can use InputStream's readFully method, which will actually read all bytes, vs. read, which reads at least 1, but is not guaranteed to read all of it. The idea is: It'll read the smallest chunk that is available. Imagine a network pipe where bytes are dribbling into the network card's buffer, one byte a second. If so far 250 bytes have dribbled in and you call .read(some500SizeByteArr), then you get 250 bytes (250 of the 500 bytes are filled in, and 250 is returned). If you call .readFully(some500SizeByteArr), then the code will wait about 250 seconds, and then returns 500, and fills in all 500 bytes. That's the difference, and that explains why read works the way it does. Said differently: If you do not check what read() is returning, your code is definitely broken.

If you do not know how much data there is, your only option involves a while loop, or to call a helper method that does that. You need to make a temporary byte array, then in a loop keep calling read until it returns -1. For every loop, take the bytes in that array from 0 to (whatever the read call returned), and send these bytes someplace else. For example, a ByteArrayOutputStream.

when I deserialize data in another runtime it changes and, for example, instead of boolean, I retrieve int

The java serialization system isn't magically changing your stuff on you. Well, put a pin that. Most likely the class file available in the first run (where you saved the blob in the db) was different vs what it looked like in your second run. Voila, problem.

More generally this is a problem in serialization. If you serialize, say, class Person {Date dob; String name;}, and then in a later version of the software you realize that using a j.u.Date to store a date of birth is a very silly idea, as Date is an unfortunately named class (it represents an instant in time and not a date at all), so you replace it with a LocalDate instead, thus ending up with class Person{LocalDate dob; String name;}, then how do you deal with the problem that you now want to deserialize a BLOB that was made back when the Person.class file still had the broken Date dob; field?

The answer is: You can't. Java's baked in serialization mechanism will flat out throw an exception here, it will not try to do this. This is the serialVersionUID system: Classes have an ID and changing anything about them (such as that field) changes this ID; the ID is stored in the serialized data. If the IDs don't match, deserialization cannot be done. You can force the ID (make a field called serialVersionUID - you can search the web for how to do that), but then you'd still get an error, java's deserializer will attempt to deserialize a Date object into a LocalDate dob; field and will of course fail.

Classes can write their own code to solve this problem. This is non-trivial and is irrelevant to you, as you're building a framework and presumably can't pop in and write code for your testing framework's userbase's custom class files.

I told you to put a pin in 'the serialization mechanism isnt going to magically change types on you'. Put in sufficient effort with overriding serialVersionUID and such and you can end up there. But that'd be because you wrote code that confuses types, e.g. in your readObject implementation (again, search the web for java's serialization mechanism, readObject/writeObject - or just start reading the javadoc of java.io.Serializable, that's a good starting-off point).

You create objects for no purpose, you seem to have some trouble with the distinction between a variable/reference and an object. You aren't using try-with-resources. The way your SELECT calls are made suggests you have an SQL injection security issue. e.printStackTrace() as line line in a catch block is always incorrect.

The exception you get SplitBrainMergeValidationOp means that the Hazelcast cluster might have been started in the split-brain and later tries to merge into one cluster. Could you check if you follow all the Hazelcast Kubernetes recommendations?

Especially, check if you use StatefulSet (not Deployment). In the case of DNS Lookup discovery, using Deployment may cause Hazelcast to start in the split-brain mode.

Since there's no way to guarantee the order of execution, it's hard to find a to write this particular function. But I can think of a few work-around for the problem you have:

1. It sounds like you are only rapidly updating right after it's created? In that case maybe what can work is having your create write to a draft collection and have your update copy the updated data into the searchable collection.
2. Rather than updating, add new documents with timestamp and filter the result at query-time.
3. If your client can tolerate stale data, you can have a scheduled function that syncs all documents with Algolia.

I work for Oxygen XML Editor and as far as I know Oxygen places the caret at that specific character offset, so we do not use the visual tab width representation for anything in this context. One thing you should be aware is that in Oxygen by default if you press the "Tab" key you actually insert 4 spaces, if you want to insert the "Tab" character instead, in the Preferences->"Editor / Format" page there is a setting named "Indent with tabs" which needs to be checked.

I would propose the following (which we use successful in a lot of Go Lambdas).

main.go

Apache Arrow utilizes a pretty similar idea of Java off-heap storage as Apache Ignite does. For Apache Arrow it means that objects like IntVector don't actually store data in their on-heap layout. They just store a reference to a buffer containing an off-heap address of a physical representation. Technically it's a long offset pointing to a chunk of memory within JVM address space.

When you restart your JVM, address space changes. But in your Apache Ignite native persistence there's a record holding an old pointer. It leads to a SIGSEGV because it's not in the JVM address anymore (in fact it doesn't even exist after a restart).

You could use Apache Arrow serialization machinery to store data permanently in Apache Ignite or even somewhere else. But in fact after that you're going to lose Apache Arrow preciousness as a fast in-memory columnar store. It was initially designed to share off-heap data across multiple data-processing solutions.

Therefore I believe that technically it could be possible to leverage Apache Ignite binary storage format. In that case a custom BinarySerializer should be implemented. After that it would be possible to use it with the Apache Arrow vector classes.

You need to add_permission to your function, so that your api gateway is allowed to invoke it. Example values for such permissions are here.