David, The problem is exactly that ResourceLifecycleStates do not correspond to specific Job statuses (JobReady condition) in most cases. Let me give you a concrete example:
ResourceLifecycleState.STABLE means that app/job defined in the spec has been successfully deployed and was observed running, and this spec is now considered to be stable (won't be rolled back). Once a resource (FlinkDeployment) reached STABLE state, it won't change unless the user changes the spec. At the same time, this doesn't really say anything about job health/readiness at any given future time. 10 minutes later the job can go in an unrecoverable failure loop and never reach a running status, the ResourceLifecycleState will remain STABLE. This is actually not a problem with the ResourceLifecycleState but more with the understanding of it. It's called ResourceLifecycleState and not JobState exactly because it refers to the upgrade/rollback/suspend etc lifecycle of the FlinkDeployment/FlinkSessionJob resource and not the underlying flink job itself. But this is a crucial detail here that we need to consider otherwise the "Ready" condition that we may create will be practically useless. This is the reason why @morh...@apache.org <morh...@apache.org> and I suggest separating this to at least 2 independent conditions. One could be the UpgradeCompleted/ReconciliationCompleted or something along these lines computed based on LifecycleState (as described in your proposal but with a different name). The other should be JobReady which could initially work based on the JobStatus.state field but ideally would be user configurable ready condition such as (job running at least 10 minutes, running and have taken checkpoints etcetc). These 2 conditions should be enough to start with and would actually provide a tangible value to users. We can probably leave out ClusterReady on a second thought. Cheers, Gyula On Wed, May 29, 2024 at 5:16 PM David Radley <david_rad...@uk.ibm.com> wrote: > Hi Gyula, > Thank you for the quick response and confirmation we need a Flip. I am not > an expert at K8s, Lajith will answer in more detail. Some questions I had > anyway: > > I assume each of the ResourceLifecycleState do have a corresponding > jobReady status. You point out some mistakes in the table, for example that > STABLE should be NotReady; thankyou. If we put a reason mentioning the > stable state, this would help us understand the jobStatus. > > I guess the jobReady is one perspective that we know is useful (with > corrected mappings from ResourceLifecycleState and with reasons). Can I > check that the 2 proposed conditions would also be useful additions? I > assume that in your proposal when jobReady is true, then UpgradeCompleted > condition would not be present and ClusterReady would always be true? I > know conditions do not need to be orthogonal, but I wanted to check what > your thoughts are. > > Kind regards, David. > > > > > From: Gyula Fóra <gyula.f...@gmail.com> > Date: Wednesday, 29 May 2024 at 15:28 > To: dev@flink.apache.org <dev@flink.apache.org> > Cc: morh...@apache.org <morh...@apache.org> > Subject: [EXTERNAL] Re: [DISCUSS] FLIP-XXX Add K8S conditions to Flink CRD > Hi David! > > This change definitely warrants a FLIP even if the code change is not huge, > there are quite some implications going forward. > > Looping in @morh...@apache.org <morh...@apache.org> for this discussion. > > I have some questions / suggestions regarding the condition's meaning and > naming. > > In your proposal you have: > - Ready (True/False) -> This condition is intended for resources which are > fully ready and operational > - Error (True) -> This condition can be used in scenarios where any > exception/error during resource reconcile process > > The problem with the above is that the implementation does not well reflect > this. ResourceLifecycleState STABLE/ROLLED_BACK does not actually mean the > job is running, it just means that the resource is fully reconciled and it > will not be rolled back (so the current pending upgrade is completed). This > is mainly a fault of the ResourceLifecycleState as it doesn't capture the > job status but one could argue that it was "designed" this way. > > I think we should probably have more condition types to capture the > difference: > - JobReady (True/False) -> Flink job is running (Basically job status but > with transition time) > - ClusterReady (True/False) -> Session / Application cluster is deployed > (Basically JM deployment status but with transition time) > - UpgradeCompleted (True/False) -> Similar to what you call Ready now > which should correspond to the STABLE/ROLLED_BACK states and mostly tracks > in-progress CR updates > > This is my best idea at the moment, not great as it feels a little > redundant with the current status fields. But maybe thats not a problem or > a way to eliminate the old fields later? > > I am not so sure of the Error status and what this means in practice. Why > do we want to track the last error in 2 places? It's already in the status. > > What do you think? > Gyula > > On Wed, May 29, 2024 at 3:55 PM David Radley <david_rad...@uk.ibm.com> > wrote: > > > Hi, > > Thanks Lajith for raising this discussion thread under the Flip title. > > > > To summarise the concerns from the other discussion thread. > > > > “ > > - I echo Gyula that including some examples and further explanations > might > > ease reader's work. With the current version, the FLIP is a bit hard to > > follow. - Will the usage of Conditions be enabled by default? Or will > there > > be any disadvantages for Flink users? If Conditions with the same type > > already exist in the Status Conditions > > > > - Do you think we should have clear rules about handling rules for how > > these Conditions should be managed, especially when multiple Conditions > of > > the same type are present? For example, resource has multiple causes for > > the same condition (e.g., Error due to network and Error due to I/O). > Then, > > overriding the old condition with the new one is not the best approach > no? > > Please correct me if I misunderstood. > > “ > > > > I see the Google doc link has been reformatted to match the Flip > template. > > > > To explicitly answer the questions from Jeyhun and Gyula: > > - “Will the usage of Conditions be enabled by default?” Yes, but this is > > just making the status content useful, whereas before it was not useful. > > - in terms of examples, I am not sure what you would like to see, the > > table Lajith provided shows the status for various > ResourceLifecycleStates. > > How the operator gets into these states is the current behaviour. The > > change just shows the appropriate corresponding high level status – that > > could be shown on the User Interfaces. > > - “will there be any disadvantages for Flink users?” None , there is just > > more information in the status, without this it is more difficult to work > > out the status of the job. > > - Multiple conditions question. The status is showing whether the job is > > ready or not, so as long as the last condition is the one that is shown - > > all is as expected. I don’t think this needs rules for precedence and the > > like. > > - The condition’s Reason is going to be more specific. > > > > Gyula and Jeyhun, is the google doc clear enough for you now? Do you feel > > you feedback has been addressed? Lajith and I are happy to provide more > > details. > > > > I wonder whether this change is big enough to warrant a Flip, as it is so > > small. We could do this in an issue. WDYT? > > > > Kind regards, David. > > > > > > From: Lajith Koova <lajith...@gmail.com> > > Date: Wednesday, 29 May 2024 at 13:41 > > To: dev@flink.apache.org <dev@flink.apache.org> > > Subject: [EXTERNAL] [DISCUSS] FLIP-XXX Add K8S conditions to Flink CRD > > Hello , > > > > > > Discussion thread here: > > https://lists.apache.org/thread/dvy8w17pyjv68c3t962w49frl9odoz4z to > > discuss a proposal to add Conditions field in the CR status of Flink > > Deployment and FlinkSessionJob. > > > > > > Note : Starting this new thread as discussion thread title has been > > modified to follow the FLIP process. > > > > > > Thank you. > > > > Unless otherwise stated above: > > > > IBM United Kingdom Limited > > Registered in England and Wales with number 741598 > > Registered office: PO Box 41, North Harbour, Portsmouth, Hants. PO6 3AU > > > > Unless otherwise stated above: > > IBM United Kingdom Limited > Registered in England and Wales with number 741598 > Registered office: PO Box 41, North Harbour, Portsmouth, Hants. PO6 3AU >
