Sounds like the new workflow plugin will suit your needs a lot better, but for now you will probably want to use the matrix project type
On 22 June 2014 10:30, <[email protected]> wrote: > Good evening everyone, > > I've heard great things about Jenkins over the years, and finally decided > to give it a serious test drive at work recently. My experiences have been > massively underwhelming thus far, and I'm hoping this is because of a > critical knowledge gap on my part regarding Jenkins' capabilities. I > welcome any pointers, insight, and recommendations. > > Our use case: > > - We need to build *22 different computational tools*. > - Our binaries are collectively comprised of *281 individual libraries* > . > - We use *60 to 70 build machines* with hardware ranging from 2-core, > 3GB of RAM to 64-core, 1TB of RAM. > > We've logically divided up our build process into several stages and tasks: > > - *Stage 1: Prepare a node for building.* 1 task (let's call it > *prepare*) to sync code, set up a temporary directory to work in, and > run parsers/generators (lex, yacc, bison, etc.). > - *Stage 2: Build a library.* 281 different tasks (let's call them > *build_lib1*, *build_lib2*, *build_lib3*, etc.) for 281 different > libraries. > - *Stage 3: Build a binary.* 22 different tasks (let's call them > *build_binA*, *build_binB*, *build_binC*, etc.) for 22 different > binaries. > > We've been able to hook Jenkins up to our SCM system without any issue, > and Jenkins begins building within seconds after an engineer makes a > commit. That part works like a charm. But as soon as we try to use Jenkins > to support any kind of complex build workflow, we run into massive > inefficiencies. Here's what we want to achieve, but I'm not certain if > Jenkins even supports this kind of behavior: > > 1. As soon as a commit is made, all available nodes enter into *Stage > 1* by running the *prepare* task. They sync, create directories, and > run generators. Our fastest nodes can complete this stage in about 50 > seconds, while others take upwards of 10 minutes. > 2. As soon as a particular node finishes Stage 1, it should > immediately enter *Stage 2* and begin running library build tasks ( > *build_lib1*, *build_lib2*, etc.). This stage can take several hours > to complete. Each node should continue performing library build tasks until > all 281 have been completed and there are no more tasks in the Stage 2 > queue. When a lib is finished building, it's copied to a network path > visible to all other nodes. > 3. The instant after Stage 2 completes, all nodes should immediately > enter *Stage 3* and begin running binary build tasks (*build_binA*, > *build_binB*, etc.). When a binary is finished building, it is > uploaded to a network path that all other binaries from this run will be > copied to. > > As best as I can tell, Jenkins either does not support the above workflow, > or requires such a complex combination of plugins that none of us at work > have been able to get anything fully working. Here are the issues we're > facing: > > 1. *How do we tell Jenkins to execute the same job (prepare) on all > nodes?* We must be missing something very fundamental here because > this is what I would consider core functionality in any build system. We > want to be able to say "Run this job (*prepare*) on all nodes that > match label XYZ", but it seems like Jenkins only runs the job once, on some > random node in that label group. In order to get execution the same job > simultaneously across all nodes, we've had to resort to some pretty > ugly hacks or weird combinations of plugins, none of which are what I would > call production-ready. > 2. *How do we tell a Jenkins node to begin Stage 2 immediately after > it finishes Stage 1?* Once our fastest nodes finish Stage 1, they're > ready to begin building libraries, and do not need to wait for the slower > nodes to finish preparing themselves. Our 64-core machines are often ready > in about 50 seconds, whereas many of our 2-core and 4-core machines take 8 > to 10 minutes to be ready. Our builds can take several hours to complete, > and if we have our entire farm sitting idle while we wait on our slowest > nodes to finish preparing themselves, we're wasting massive amounts of > resources. > 3. How do we tell a Jenkins node to begin Stage 3 *immediately** after > all tasks in Stage 2 have completed?* Once tasks in Stage 2 complete, > we should be clear to begin the tasks in Stage 3. How do we ensure we're > properly gated here? There are some plugins out there that appear to do > just that, but they're either marked as deprecated or no longer supported. > > So how does everyone here use Jenkins? I consider our use case to be > ridiculously common among large projects, but it seems almost like Jenkins > wasn't built for the above kind of workflow. We've now started looking at > plugins to provide the above functionality, but installing a large number > of different plugins - all with different documentation standards and > development histories, some of which seem to be abandoned - just to get > things working all seems so... fragile. > > Are we doing this all wrong? Does Jenkins not support this kind of > functionality out of the box? > > I welcome your input and feedback, > -JJ > > -- > You received this message because you are subscribed to the Google Groups > "Jenkins Users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "Jenkins Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
