Thanks for moving this forwards Ekaterina. I think what we perhaps discovered is that there’s not really any consensus about how to best do config files. I think in this situation it’s best to defer to the one who’s actually putting in the time to _do_, so I am more than happy to defer to your decisions.
I’m sure everyone is looking forward to the improved consistency of this work. From: Ekaterina Dimitrova <e.dimitr...@gmail.com> Date: Wednesday, 20 October 2021 at 22:27 To: dev@cassandra.apache.org <dev@cassandra.apache.org> Subject: Re: [DISCUSS] CASSANDRA-15234 Hi everyone, I think it is time to summarize the discussion. First of all, thank you for all the valuable input, suggestions, concerns, and comments! The things that I believe we all agree on: - Simplicity for maintenance on our end - automation as much as possible so we don’t have to maintain more than one configuration file and our config is less prone to human errors while adding new features - Simplicity for our users - as less confusing and as simple as possible and having in mind the users’ toolset - Simplicity for testing and verification of the different config file formats It seems to me that most people want to see committed both proposed versions(feel free to correct me if I am wrong) with revision of the default values and potentially commented out all parameters that are not really mandatory to be changed. Also, versions with striped comments plus a way to maintain everything automatically, as much as possible. With that said it seems to me the current patch in CASSANDRA-15234 can be committed after rebase and addressing any outstanding review comments. The new version of cassandra.yaml, grouping the parameters can be added in a new ticket by me or anyone with free cycles for that. It will require additional work on the backward compatibility and the opportunity for Cassandra to operate on all of the current versions but it will be new additional opportunity which doesn’t disqualify the old ones so it seems as a fair game to be added at any point in time in the future as it won’t be a breaking change. We won’t replace anything. We will only add more options. If someone disagrees and wants to implement all possible options and functionalities at once, I will be happy to handover the work and try to find the time to provide feedback/reviews later. Please do not hesitate to correct me if I misunderstood something. I will leave this discussion open until Monday and if there are no objections I will continue with CASSANDRA-15234 as per my proposal. Best regards, Ekaterina On Fri, 10 Sep 2021 at 20:18, Patrick McFadin <pmcfa...@gmail.com> wrote: > Ah, I feel like cassandra.yaml discussions are such an evergreen topic. > > This was something brought up a while back, but I remember years ago we > talked about emulating the config options that some other databases have > done. Providing different versions of the config for different approaches. > For instance, MySQL has had 'my-small.cnf' with just the bare minimum > config and restricted parameters for something like a laptop. A friendly > option for newcomers would be a clearly labeled 'cassandra-small.yaml' > with just the bare minimum and good comments. Then people new to Cassandra > wouldn't have a panic moment wondering if they have to know what concurrent > compactors are and how many you actually need? (Is there a right answer > even???) It's tackling the way operators approach config by the use case > they are trying to satisfy. Run one node on my laptop. Run a small cluster > on a budget cloud server. Run any size cluster on a ginormous server. > > Unfortunately, the cleaner solution would be how Apache HTTD solved it back > in the day with include files. It made config management much easier and > the overwhelm factor much lower. Yaml doesn't support it and it would all > have to be custom code in the Cassandra config loader. Not the best option > really. > > Back to the original question, I think Ekaterina's sectioned version could > be used for new operators because there is a lot to learn looking at the > comments. Publish the following options: > > cassandra-small.yaml: Just the 'Quickstart' section > cassandra-medium.yaml: 'Quickstart' and 'Commonly used' with sane defaults > cassandra-advanced.yaml: Every section > > The addition is a similarly named JVM properties file . > > As somebody who has been using Cassandra for a while and would like to have > a more verbose version (especially for config management) Benedict's > grouped version is fantastic. Just one option there: > > cassandra-full.yaml > > That's my idea to satisfy the various operators that approach a new > install. > > Patrick > > On Fri, Sep 10, 2021 at 3:31 PM Jeremiah D Jordan < > jeremiah.jor...@gmail.com> > wrote: > > > > Also, if you run the above command you will see we actually have a lot > > of things show (129 lines)… it would be nice to clean it up as only a > small > > subset is required and most shown normal users won’t care > > > > +1 for this. It would be good to clean up the config code and yaml such > > that only “things that are required to be changed” are not commented out > in > > the file, and everything else is commented out by default. Last I > checked > > there were many fields that when commented out would not use a sensible > > value, or would result in NPE’s because they didn’t have a code level > > default. > > > > -Jeremiah > > > > > On Sep 10, 2021, at 1:24 PM, David Capwell <dcapw...@apple.com.INVALID > > > > wrote: > > > > > > We can have both, but I would hope we do not have humans maintaining > > both. If we maintain the commented one, and did something like the below > > while we compile then the burden to maintain doesn’t exist > > > > > > # remove comments and empty lines > > > $ egrep -v '^[[:space:]]*#|^[[:space:]]*$' conf/cassandra.yaml.doc > > > conf/cassandra.yaml > > > > > > We do this right now with conf/hotspot_compiler so as long as our build > > maintains the other file +1 > > > > > > Also, if you run the above command you will see we actually have a lot > > of things show (129 lines)… it would be nice to clean it up as only a > small > > subset is required and most shown normal users won’t care > > > > > >> On Sep 3, 2021, at 6:45 AM, bened...@apache.org wrote: > > >> > > >>> I think as the comments were stripped only for the POC. I guess many > > of them will get back > > >> in the actual doc version unfortunately. > > >> > > >> Well, I think the grouped format lends itself to much briefer > comments, > > with groups of related parameters getting an overall description. Even > as a > > developer who understands most of the toggles I found the old file very > > hard to navigate. > > >> > > >> I also don’t see why we cannot have both heavily commented versions > and > > uncommented (or lightly commented) versions. > > >> > > >> I don’t personally see why multiple different config templates would > be > > confusing if they’re in a suitably labelled directory, even if we settle > on > > one for the default. It might even be nice to have a pared-down config > that > > has only those properties we expect the normal user to need, so it’s > > particularly easy to navigate. > > >> > > >> > > >> From: Ekaterina Dimitrova <e.dimitr...@gmail.com> > > >> Date: Friday, 3 September 2021 at 14:40 > > >> To: dev@cassandra.apache.org <dev@cassandra.apache.org> > > >> Subject: Re: [DISCUSS] CASSANDRA-15234 > > >>>> > > >>>> It’s worth noting that the two don’t have to be in >conflict: we > could > > >>> offer two template yaml with the parameters grouped differently, for > > users > > >>> to decide for themselves. > > >> > > >> Sure, my only concern is that three versions of the yaml could bring > > >> confusion (we will have backward compatibility to the current one for > > some > > >> time). But it might be only me. I am open for feedback > > >> > > >> > > >>> If we can document this, it would be great as stuff >like “enabled” > are > > >>> inconsistent so not sure if I did it properly =D > > >>> > > >> Well, this is for now only in the ticket in the first version but no > one > > >> raised any concern. We will definitely have to update our docs on this > > and > > >> whatever else we came to agreement on - both for users and > contributors. > > >> > > >>> though I will agree that it can be hard for some >tools (such > > >>> as bash templating), but feel we can always find a >common ground > > >> Valid point and I believe it is one of the reasons we delayed the > > ticket, > > >> in order to get feedback on that. I am really interested to hear what > > >> concerns people might have. > > >> > > >> > > >>> Opening up a 1500+ line .yaml file is very daunting, >even if most of > > it is > > >>> comments. Can't blame folks for being >overwhelmed at the prospect of > > >> tuning > > >>> Cassandra w/that as our operator config API. :) > > >> I am all in for simplification and to make our users’ lives easier. > But > > at > > >> this point we shouldn’t be comparing the length of the files I think > as > > the > > >> comments were stripped only for the POC. I guess many of them will get > > back > > >> in the actual doc version unfortunately. > > >> > > >> Thank you all, > > >> Ekaterina > > >> > > >> On Thu, 2 Sep 2021 at 20:07, Joshua McKenzie <jmcken...@apache.org> > > wrote: > > >> > > >>> Reading through the two, the grouping approach seems like it's a lot > > more > > >>> friendly to newcomers as well as providing context specific cues for > > >>> relationships between params you're editing. Showing and not telling, > > if > > >>> you will. > > >>> > > >>> Opening up a 1500+ line .yaml file is very daunting, even if most of > > it is > > >>> comments. Can't blame folks for being overwhelmed at the prospect of > > tuning > > >>> Cassandra w/that as our operator config API. :) > > >>> > > >>> ~Josh > > >>> > > >>> On Thu, Sep 2, 2021 at 1:48 PM David Capwell > > <dcapw...@apple.com.invalid> > > >>> wrote: > > >>> > > >>>> Thanks for bringing this back up; Caleb and I were talking about the > > lack > > >>>> of clarity with regard to CASSANDRA-16896, fleshing this out would > > make > > >>>> those configs nicer! > > >>>> > > >>>>> To standardize naming - that we did by agreeing to the form > noun_verb > > >>>> > > >>>> If we can document this, it would be great as stuff like “enabled” > are > > >>>> inconsistent so not sure if I did it properly =D > > >>>> > > >>>>> > > >>>>> Provision of values with units while maintaining backward > > >>>> compatibility. > > >>>> > > >>>> +1000000000000 > > >>>> > > >>>> I really hate local_read_size_threshold_kb; I would love > > >>>> local_read_size_threshold: 10kb. Once we have the infrastructure in > > >>> place > > >>>> (believe your patch before had these tools) I would love to switch! > > >>>> > > >>>> > > >>>>> Another proposal is done by Benedict; grouping the config > parameters. > > >>>> > > >>>> Yep, this is what triggered Caleb and I to talk about this thread! > To > > >>>> group or not to group; that is the question > > >>>> > > >>>> Personally I like grouping from an organization point of view so am > in > > >>>> favor of that; though I will agree that it can be hard for some > tools > > >>> (such > > >>>> as bash templating), but feel we can always find a common ground > > >>>> > > >>>> > > >>>>> On Sep 2, 2021, at 8:44 AM, bened...@apache.org wrote: > > >>>>> > > >>>>> Thanks for bringing this to the list Ekaterina! > > >>>>> > > >>>>> It’s worth noting that the two don’t have to be in conflict: we > could > > >>>> offer two template yaml with the parameters grouped differently, for > > >>> users > > >>>> to decide for themselves. > > >>>>> > > >>>>> The proposals primarily define parameter names differently, with my > > >>>> proposal going by kind->place, and the other proposal maintaining > > >>> (mostly) > > >>>> the existing name form (which is a bit more like place->kind). While > > the > > >>>> example yaml groups by kind, you can convert nested definitions > into a > > >>>> ‘dot’ form (e.g. limits.concurrency.reads) for use in a different > > >>> grouping. > > >>>>> > > >>>>> One advantage of grouping parameters together is that it aids > > >>>> maintaining coherency of naming between systems, and also > potentially > > >>>> permits a more succinct config file and better discovery. But it’s > far > > >>> from > > >>>> a silver bullet, as value judgements have to be made about where the > > >>>> grouping lines are. I’m sure anything we settle on will be a huge > > >>>> improvement over the status quo, however. > > >>>>> > > >>>>> > > >>>>> > > >>>>> > > >>>>> From: Ekaterina Dimitrova <e.dimitr...@gmail.com> > > >>>>> Date: Thursday, 2 September 2021 at 16:32 > > >>>>> To: dev@cassandra.apache.org <dev@cassandra.apache.org> > > >>>>> Subject: [DISCUSS] CASSANDRA-15234 > > >>>>> Hi team, > > >>>>> > > >>>>> I would like to bring to the attention of the community > > >>> CASSANDRA-15234, > > >>>>> standardise config and JVM parameters. > > >>>>> > > >>>>> This is work we discussed back in Summer 2020 just before our first > > 4.0 > > >>>>> Beta release. During the discussion we figured out that there is > more > > >>>> than > > >>>>> one option to do the job and not enough time to get user feedback > and > > >>>>> finish it so this was delayed post-4.0 And here I am, bringing it > > back > > >>> to > > >>>>> the table. > > >>>>> > > >>>>> This work’s goal is: > > >>>>> > > >>>>> - > > >>>>> > > >>>>> To standardize naming - that we did by agreeing to the form > noun_verb > > >>>>> - > > >>>>> > > >>>>> Provision of values with units while maintaining backward > > >>>> compatibility. > > >>>>> > > >>>>> > > >>>>> Those two parts are more or less already done. > > >>>>> > > >>>>> More interesting is the third part - reorganizing the > cassandra.yaml > > >>>> file. > > >>>>> > > >>>>> My personal approach was to split it into sections, done here > > >>>>> < > > >>>> > > >>> > > > https://github.com/ekaterinadimitrova2/cassandra/blob/b4eebe080835da79d032f9314262c268b71172a8/conf/cassandra.yaml > > >>>>> > > >>>>> . > > >>>>> > > >>>>> Another proposal is done by Benedict; grouping the config > parameters. > > >>>>> > > >>>>> To make it clearer, he created a yaml > > >>>>> < > > >>>> > > >>> > > > https://github.com/belliottsmith/cassandra/blob/5f80d1c0d38873b7a27dc137656d8b81f8e6bbd7/conf/cassandra_nocomment.yaml > > >>>>> > > >>>>> with comments mostly stripped. > > >>>>> > > >>>>> In his version, there are basic settings for network, disk etc all > > >>>> grouped > > >>>>> together, followed by operator tuneables mostly under limits within > > >>> which > > >>>>> we now have throughput, concurrency, capacity. This leads to > settings > > >>> for > > >>>>> some features being kept separate (most notably for caching), but > > helps > > >>>> the > > >>>>> operator understand what they have to play with for controlling > > >>> resource > > >>>>> consumption. > > >>>>> > > >>>>> I am interested to hear what people think about the two options or > if > > >>>>> anyone has another idea to share, open discussion. > > >>>>> > > >>>>> Thank you, > > >>>>> > > >>>>> Ekaterina > > >>>> > > >>>> > > >>>> > --------------------------------------------------------------------- > > >>>> To unsubscribe, e-mail: dev-unsubscr...@cassandra.apache.org > > >>>> For additional commands, e-mail: dev-h...@cassandra.apache.org > > >>>> > > >>>> > > >>> > > > > > > > > > --------------------------------------------------------------------- > > > To unsubscribe, e-mail: dev-unsubscr...@cassandra.apache.org > > > For additional commands, e-mail: dev-h...@cassandra.apache.org > > > > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: dev-unsubscr...@cassandra.apache.org > > For additional commands, e-mail: dev-h...@cassandra.apache.org > > > > >
