I have two questions-
Q1. Is there a more efficient way to handle the error situation in MPI, other than check-point/rollback? I see that if a node "dies", the program halts abruptly.. Is there any way to go ahead with the execution after a node dies ?? (no issues if it is at the cost of accuracy)
Q2. I read in "http://stackoverflow.com/questions/144309/what-is-the-best-mpi-implementation", that OpenMPI has better fault tolerance and recently MPICH-2 has also come up with similar features.. does anybody know what they are and how to use them? is it a "mode"? can they help in the situation stated in Q1 ?
kindly reply. Thank you.
MPI - all implementations - have had the ability to continue after an error for a while. The default is to die - that is, the default error handler is MPI_ERRORS_ARE_FATAL - but that can be set (eg, see the discussion here). But the standard doesn't currently much beyond that; that is, it's hard to recover and continue after such an error. If your program is sufficiently simple - some sort of master-worker type of setup - it may be possible to continue this way.
The MPI forum is currently working on what will become MPI-3, and error handling and fault tolerance will be an important component of the new standard (there's a working group dedicated to the topic). Until that work is complete, however, the only way to get stronger fault tolerance out of MPI is to use earlier, nonstandard, extensions. FT-MPI was a project that developed a very robust MPI, but unfortuantely it's based on MPI1.2; a very early version of the standard. The claim here is that they're now working with OpenMPI, but I don't know what's become of that. There's MPICH-V, based on MPI2, but that's more checkpoint-restart based than what I think you're looking for.
Updated to add: The fault tolerance didn't make it into MPI-3, but the working group continues its work and the expectation is that something will result from that before too long.
Related
I am creating this new topic because I am using the OpenMDAO platform, and more specifically its design of experiment option. I would like to know if there is a proper way to interrupt and stop the computations if a condition is met in my program.
I have already used OpenMDAO optimizers to study and solve some problems and to stop the computations I used to raise an Exception to stop the program. This strategy seems to work for optimizers but not so much when I am using the LatinHypercubeGenerator driver: it is like the OpenMDAO program is still trying to compute the points even if Exception or RuntimeError are raise within the OpenMDAO explicit component function "compute".
In that respect I am wondering if there is a way to kill OpenMDAO during calculations. I tried to check if an OpenMDAO built-in attribute or method could do the job, but I have not found anything.
Does anyone know how to stop OpenMDAO DOE computations?
Many thanks in advance for any advice/help
As of OpenMDAO V3.18, there is no way to add some kind off a stopping condition to the DOE driver. You mention using AnalysisError to achieve this for other optimizers. This won't work in general either, since some drivers will intentionally catch those errors, react, and attempt to keep running the optimization.
You can see the run code of the driver, where a for loop is made and some try/catch blocks are used to record the success/failure of specific cases.
My suggestion for achieving what you want would be to copy the driver code into your model directory and make your own custom drivers. You can add whatever kind of termination condition you like, either based on results of a single case or some statistical analysis of the currently run cases.
If you come up with a clean way of doing it, you can always submit a POEM and/or a pull request to propose adding your new functionality to the mainline of OpenMDAO.
My girlfriend was asked the below question in an interview:
We trigger 5 independent APIs simultaneously. Once they have all completed, we want to trigger a function. How will you design a system to do this?
My girlfriend replied she will use a flag variable, but the interviewer was evidently not happy with it.
So, is there a good way in which this could be handled (in a distributed context)? Note that each of the 5 API calls are made by different servers and the function to be triggered is on a 6th server.
The other answers suggesting Promises seem to assume all these requests necessarily come from the same client. If the context here is distributed systems, as you said it is, then I don't think those are valid answers. If they were, then the interview question would have nothing to do with distributed systems, except to essay your girlfriend's ability to recognize something that isn't really a distributed systems problem.
And the question does have the shape of some classic problems in distributed systems. It sounds a lot like YouTube view counting: How do you achieve qualities like atomicity and consistency in a multi-threaded, multi-process, or multi-client environment? Failing to recognize this, thinking the answer could be as simple as "a flag", betrayed a lack of experience in distributed systems.
Another thing about that answer is that it leaves many ambiguities. Where does the flag live? As a variable in another (Java?) API? In a database? In a file? Even in a non-distributed context, these are important questions. And if she had gone on to address these questions, even being innocent of all the distributed systems complications, she might have happily fallen into a discussion of the kinds of D.S. problems that occur when you use, say, a file; and how using a ACID-compliant database might solve those problems, and what the tradeoffs might be there... And she might have corrected herself and said "counter" instead of "flag"!
If I were asked this, my first thought would be to use promises/futures. The idea behind them is that you can execute time-consuming operations asynchronously and they will somehow notify you when they've completed, either successfully or unsuccessfully, typically by calling a callback function. So the first step is to spawn five asynchronous tasks and get five promises.
Then I would join the five promises together, creating a unified promise that represents the five separate tasks. In JavaScript I might call Promise.all(); in Java I would use CompletableFuture.allOf().
I would want to make sure to handle both success and failure. The combined promise should succeed if all of the API calls succeed and fail if any of them fail. If any fail there should be appropriate error handling/reporting. What happens if multiple calls fail? How would a mix of successes and failures be reported? These would be design points to mention, though not necessarily solve during the interview.
Promises and futures typically have modular layering system that would allow edge cases like timeouts to be handled by chaining handlers together. If done right, timeouts could become just another error condition that would be naturally handled by the error handling already in place.
This solution would not require any state to be shared across threads, so I would not have to worry about mutexes or deadlocks or other thread synchronization problems.
She said she would use a flag variable to keep track of the number of API calls have returned.
One thing that makes great interviewees stand out is their ability to anticipate follow-up questions and explain details before they are asked. The best answers are fully fleshed out. They demonstrate that one has thought through one's answer in detail, and they have minimal handwaving.
When I read the above I have a slew of follow-up questions:
How will she know when each API call has returned? Is she waiting for a function call to return, a callback to be called, an event to be fired, or a promise to complete?
How is she causing all of the API calls to be executed concurrently? Is there multithreading, a fork-join pool, multiprocessing, or asynchronous execution?
Flag variables are booleans. Is she really using a flag, or does she mean a counter?
What is the variable tracking and what code is updating it?
What is monitoring the variable, what condition is it checking, and what's it doing when the condition is reached?
If using multithreading, how is she handling synchronization?
How will she handle edge cases such API calls failing, or timing out?
A flag variable might lead to a workable solution or it might lead nowhere. The only way an interviewer will know which it is is if she thinks about and proactively discusses these various questions. Otherwise, the interviewer will have to pepper her with follow-up questions, and will likely lower their evaluation of her.
When I interview people, my mental grades are something like:
S — Solution works and they addressed all issues without prompting.
A — Solution works, follow-up questions answered satisfactorily.
B — Solution works, explained well, but there's a better solution that more experienced devs would find.
C — What they said is okay, but their depth of knowledge is lacking.
F — Their answer is flat out incorrect, or getting them to explain their answer was like pulling teeth.
I am learning FP and got introduced to the concept of property-based testing and for someone from OOP world PBT looks both useful and dangerous. It does check a lot of options, but what if there is one (or some) options that fail, but they didn't fail during your first let's say Jenkins build. Then next time you run the build the test may or may not fail, doesn't it kill the entire idea of repeatable builds?
I see that some people explored options to make the tests deterministic, but then if such test doesn't catch an error it will never catch it.
So what's better approach here? Do we sacrifice build repeatability to eventually uncover a bug or do we take the risk of never uncovering it, but get our repeatability back?
(I hope that I properly understood the concept of PBT, but if I didn't I would appreciate if somebody could point out my misconceptions)
Doing a lot of property-based testing I don’t see indeterminism as a big problem. I basically experience three types of it:
A property is really indeterministic b/c some external factor - e.g. timeout, delay, db config - makes it so. Those flaky tests also show up in example-based testing and should be eliminated by making the external factor deterministic.
A property fails rarely because the triggering condition is only sometimes met by pseudo random data generation. Most PBT libraries have ways to reproduce those failing runs, eg by re-using the random seed of the failing test run or even remembering the exact constellation in a database of some sort. Those failures reveal problems and are one of the reasons why we’re doing random test cases generation in the first place.
Coverage assertions („this condition will be hit in at least 5 percent of all cases“) may fail from time to time even though they are generally true. This can be mitigated by raising the number of tries. Some libs, eg quickcheck, do their own calculation of how many tries are needed to prove/disprove coverage assumptions and thereby mostly eliminate those false positives.
The important thing is to always follow up on flaky failures and find the bug, the indeterministic external factor or the wrong assumption in the property‘s invariant. When you do that, sporadic failures will occur less and less often. My personal experience is mostly with jqwik but other people have been telling me similar stories.
You can have both non-determinism and reproducible builds by generating the randomness outside the build process. You could generate it during development or during external testing.
One example would be to seed your property based tests, and to automatically modify this seed on commit. You're still making a tradeoff. A developer could be alerted of a bug unrelated to what they're working on, and you lose some test capacity since the tests might change less often.
You can tip the tradeoff further in the deterministic direction by making the seed change less often. You could for example have one seed for each program component or file, and only change it when a related file is committed.
A different approach would be to not change the seed during development at all. You would instead have automatic QA doing periodic or continuous testing with random seeds and use them to generate bug reports/issues that can be dealt with when convenient.
johanneslink's analysis of non-determinism is spot on.
There's one thing I would like to add: non-determinism is not only a rare and small cost, it's also beneficial. If the first run of your test suite is successful, insisting on determinism means insisting that future runs (of the same suite against the same system) will find zero bugs.
Usually most test suites contain many independent tests of many independent system parts, and commits rarely change large parts of the system. So even across commits, most tests test exactly the same thing before and after, where once again determinism guarantees that you will find zero bugs.
Allowing for randomness means every run has at least a chance of discovering a bug.
That of course raises the question of regression tests. I think the standard argument is something like this: to maximize value per effort you should focus your testing on the most bug-prone parts of the code. Having observed a bug in the past provides evidence about which part of the code is buggy (and which kind of bug it's likely to have). You should use that evidence to guide your testing effort. (Often with a laser-like focus on one concrete bug.)
I think this is a very reasonable argument. I also think there's more than one way of making good use of the evidence provided by bugs.
For example, you might write a generator which produces data of the same kind and shape as the data which triggered the bug the first time, and/or which is tailor made to trigger the bug.
And/or, you might want to write tests verifying specifically those properties that were violated by the buggy behavior.
If you want to judge how good these tests are, I recommend running them a couple of times (on normally sized input batches). If they trigger the bug every time, it's likely to do so in the future also.
Here's a (hopefully thought-)provoking question: is it worse to release software which has a bug it has had before, or release software with new bugs? In other words: is catching past bugs more important than catching new ones—or do do it primarily because it's easier?
If you think we do it in part because it's easier, then I don't think it matters that re-catching the bug is probabilistic: what you should really care about is something like the average bug-catching abilities of property testing—its benefits elsewhere should outweigh the fairly small chance that an old bug squeaks through, even though it got caught in (say) 5 consecutive runs of the tests when you evaluated your regression tests.
Now, if you can't reliably generate random inputs that trigger the bug even though you understand the bug just fine, or the generator which does it is large and complicated and thus costly to maintain, hand-picking a regression example seems like a perfectly reasonable choice.
I would like to know Why we need to compile the program of progress 4GL? Really what is happening behind there? Why we are getting .r file after compiled the program? When we check the syntax if its correct then we will get one message box 'Syntax is correct' how its finding the errors and showing the messages.Any explanations welcome and appreciated.
Benefits of compiled r-code include:
Syntax checking
Faster execution (r-code executes faster)
Security (r-code is not "human readable" and tampering with it will likely be noticed)
Licensing (r-code runtime licenses are much less expensive)
For "how its finding the errors and showing the messages" -- at a high level it is like any compiler. It evaluates the provided source against a syntax tree and lets you know when you violate the rules. Compiler design and construction is a fairly advanced topic that probably isn't going to fit into a simple SO question -- but if you had something more specific that could stand on its own as a question someone might be able to help.
The short answer is that when you compile, you're translating your program to a language the machine understands. You're asking two different questions here, so let me give you a simple answer to the first: you don't NEED to compile if you're the only one using the program, for example. But in order to have your program optimized (since it's already at the machine language level) and guarantee no one is messing with your logic, we compile the code and usually don't allow regular users to access the source code.
The second question, how does the syntax checker work, I believe it would be better for you to Google and choose some articles to read about compilers. They're complex, but in a nutshell what they do is take what Progress expects as full, operational commands, and compare to what you do. For example, if you do a
Find first customer where customer.active = yes no-error.
Progress will check if customer is a table, if customer.active is a field in that table, if it's the logical type, since you are filtering if it is yes, and if your whole conditions can be translated to one single true or false Boolean value. It goes on to check if you specified a lock (and default to shared if you haven't, like in my example, which is a no-no, by the way), what happens if there are multiple records (since I said first, then get just the first one) and finally what happens if it fails. If you check the find statement, there are more options to customize it, and the compiler will simply compare your use of the statement to what Progress can have for it. And collect all errors if it can't. That's why sometimes compilers will give you generic messages. Since they don't know what you're trying to do, all they can do is tell you what's basically wrong with what you wrote.
Hope this helps you understand.
During software development, there may be bugs in the codebase which are known issues. These bugs will cause the regression/unit tests to fail, if the tests have been written well.
There is constant debate in our teams about how failing tests should be managed:
Comment out failing test cases with a REVISIT or TODO comment.
Advantage: We will always know when a new defect has been introduced, and not one we are already aware of.
Disadvantage: May forget to REVISIT the commented-out test case, meaning that the defect could slip through the cracks.
Leave the test cases failing.
Advantage: Will not forget to fix the defects, as the script failures will constantly reminding you that a defect is present.
Disadvantage: Difficult to detect when a new defect is introduced, due to failure noise.
I'd like to explore what the best practices are in this regard. Personally, I think a tri-state solution is the best for determining whether a script is passing. For example when you run a script, you could see the following:
Percentage passed: 75%
Percentage failed (expected): 20%
Percentage failed (unexpected): 5%
You would basically mark any test cases which you expect to fail (due to some defect) with some metadata. This ensures you still see the failure result at the end of the test, but immediately know if there is a new failure which you weren't expecting. This appears to take the best parts of the 2 proposals above.
Does anyone have any best practices for managing this?
I would leave your test cases in. In my experience, commenting out code with something like
// TODO: fix test case
is akin to doing:
// HAHA: you'll never revisit me
In all seriousness, as you get closer to shipping, the desire to revisit TODO's in code tends to fade, especially with things like unit tests because you are concentrating on fixing other parts of the code.
Leave the tests in perhaps with your "tri-state" solution. Howeveer, I would strongly encourage fixing those cases ASAP. My problem with constant reminders is that after people see them, they tend to gloss over them and say "oh yeah, we get those errors all the time..."
Case in point -- in some of our code, we have introduced the idea of "skippable asserts" -- asserts which are there to let you know there is a problem, but allow our testers to move past them on into the rest of the code. We've come to find out that QA started saying things like "oh yeah, we get that assert all the time and we were told it was skippable" and bugs didn't get reported.
I guess what I'm suggesting is that there is another alternative, which is to fix the bugs that your test cases find immediately. There may be practical reasons not to do so, but getting in that habit now could be more beneficial in the long run.
Fix the bug right away.
If it's too complex to do right away, it's probably too large a unit for unit testing.
Lose the unit test, and put the defect in your bug database. That way it has visibility, can be prioritized, etc.
I generally work in Perl and Perl's Test::* modules allow you to insert TODO blocks:
TODO: {
local $TODO = "This has not been implemented yet."
# Tests expected to fail go here
}
In the detailed output of the test run, the message in $TODO is appended to the pass/fail report for each test in the TODO block, so as to explain why it was expected to fail. For the summary of test results, all TODO tests are treated as having succeeded, but, if any actually return a successful result, the summary will also count those up and report the number of tests which unexpectedly succeeded.
My recommendation, then, would be to find a testing tool which has similar capabilities. (Or just use Perl for your testing, even if the code being tested is in another language...)
We did the following: Put a hierarchy on the tests.
Example: You have to test 3 things.
Test the login (login, retrieve the user name, get the "last login date" or something familiar etc.)
Test the database retrieval (search for a given "schnitzelmitkartoffelsalat" - tag, search the latest tags)
Test web services (connect, get the version number, retrieve simple data, retrieve detailed data, change data)
Every testing point has subpoints, as stated in brackets. We split these hierarchical. Take the last example:
3. Connect to a web service
...
3.1. Get the version number
...
3.2. Data:
3.2.1. Get the version number
3.2.2. Retrieve simple data
3.2.3. Retrieve detailed data
3.2.4. Change data
If a point fails (while developing) give one exact error message. I.e. 3.2.2. failed. Then the testing unit will not execute the tests for 3.2.3. and 3.2.4. . This way you get one (exact) error message: "3.2.2 failed". Thus leaving the programmer to solve that problem (first) and not handle 3.2.3. and 3.2.4. because this would not work out.
That helped a lot to clarify the problem and to make clear what has to be done at first.
I tend to leave these in, with an Ignore attribute (this is using NUnit) - the test is mentioned in the test run output, so it's visible, hopefully meaning we won't forget it. Consider adding the issue/ticket ID in the "ignore" message. That way it will be resolved when the underlying problem is considered to be ripe - it'd be nice to fix failing tests right away, but sometimes small bugs have to wait until the time is right.
I've considered the Explicit attribute, which has the advantage of being able to be run without a recompile, but it doesn't take a "reason" argument, and in the version of NUnit we run, the test doesn't show up in the output as unrun.
I think you need a TODO watcher that produces the "TODO" comments from the code base. The TODO is your test metadata. It's one line in front of the known failure message and very easy to correlate.
TODO's are good. Use them. Actively management them by actually putting them into the backlog on a regular basis.
#5 on Joel's "12 Steps to Better Code" is fixing bugs before you write new code:
When you have a bug in your code that you see the first time you try to run it, you will be able to fix it in no time at all, because all the code is still fresh in your mind.
If you find a bug in some code that you wrote a few days ago, it will take you a while to hunt it down, but when you reread the code you wrote, you'll remember everything and you'll be able to fix the bug in a reasonable amount of time.
But if you find a bug in code that you wrote a few months ago, you'll probably have forgotten a lot of things about that code, and it's much harder to fix. By that time you may be fixing somebody else's code, and they may be in Aruba on vacation, in which case, fixing the bug is like science: you have to be slow, methodical, and meticulous, and you can't be sure how long it will take to discover the cure.
And if you find a bug in code that has already shipped, you're going to incur incredible expense getting it fixed.
But if you really want to ignore failing tests, use the [Ignore] attribute or its equivalent in whatever test framework you use. In MbUnit's HTML output, ignored tests are displayed in yellow, compared to the red of failing tests. This lets you easily notice a newly-failing test, but you won't lose track of the known-failing tests.