Skip to main content

Notice: this Wiki will be going read only early in 2024 and edits will no longer be possible. Please see: https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/wikis/Wiki-shutdown-plan for the plan.

Jump to: navigation, search

SimRel/Simultaneous Release Cycle FAQ

< SimRel
Revision as of 17:05, 16 December 2016 by David williams.acm.org (Talk | contribs) (Beginning "Common Errors" section)

Contents

Introduction

This page is to document answers to frequently asked questions about the yearly Simultaneous Release process or build.

How do we join the simultaneous release train

Pretty much any project that wants to join the yearly release train can and are encouraged to. Of course, it does sometimes take more work ... it is not just a matter of timing. But, often makes things much easier for your adopters and users. You should discuss with your mentors and PMC if you have any doubts or to make sure they agree its appropriate for your project.

For "how to join", see Simultaneous Release Requirements document.

How to decide if offset is +N category

It depends on who you depend on and who depends on you. The Eclipse Platform is +0, so nearly everyone else is +1, +2, or +3. For example, if you highly depended on "webtools" (which is +2) then you'd have to be a +3.

If no one else depends on you, in release train, then +3 would be a good choice. Everyone else is somewhere in between.

Of course, the lines are not always clear and pure ... such as part of emf is required at +0 but part of it is later at +1. In such cases, projects usually mark themselves as +1 and "work out the details", project-to-project, with those that need things at +0. Another example, part of a project such as WTP might depend on part of DTP and DTP might depend on part of WTP, but both need to be +2 to "fit in". In cases such, the projects might have to work out plans or agreements about having the sensitive, overlapping parts to be stable and finished early (in both API and version numbers) so neither has to build against the literal final delivery of the other.

Keep in mind, the +N category means the last possible time to drop (without notifying others). You are welcome and encouraged to have a "warm-up drop" a week or so earlier, with your "near final" bits, just to see if anything breaks or effects others, even though your final delivery may not be until +N.

What are the most important URLs

  • SimRel, the wiki category page for the current Simultaneous Release stream.

Where can the common repository be tested, before it is rolled out for a milestone or release?

The best place is the "staging area", by adding this URL to your "available software sites" list.

http://download.eclipse.org/releases/staging

Note: to get the best test, disable all other repositories in your list, or you might end up pulling something from some other repository, not staging. Be aware that moving a specific build to "staging" may happen only every few days (if you need something promoted more urgently, just ask on cross-project list).

What is the staging repository?

Conceptually, it is simply a place to hold a repository temporarily ... until the repository is promoted to a released location.

What is staging, for the current, most forward looking, yet-to-be-released stream?

 http://download.eclipse.org/staging/<trainName>

As of this writing (5/2016) it is

http://download.eclipse.org/staging/neon


Before the actual yearly release, a common milestone aggregation build is moved from 'staging' to the 'releases' area. For example, once, for each final milestone, staging is moved to

http://download.eclipse.org/releases/neon/

What is staging, for the maintenance stream?

The same as it was. That is, we used to have a URL with 'maintenance' in its path, but after Mars.2 (and prior to the Neon release) we changed so that we have a 'staging' URL for each "release train".

So once "oxygen" begins in July, then

 http://download.eclipse.org/staging/neon 

will be used to aggregate the update release (previously called "maintenance" release or SR, but now called an update release).

and the 'oxygen' aggregation will be staged in

 http://download.eclipse.org/staging/oxygen

What's the best way to test with the staging repository?

There are a couple of good tests to do, before a staging repository is released:

Test staging all by itself

Just add the staging repository to the available software sites, using preferences, and ... very important ... make sure all other sites are disabled. This then gives you the best view that everything in that one repository is correct, and all dependencies can indeed be found, and things show up in categories as expected.

After confirming the categories are as expected, usually the next test is just to install things, fresh, but often it is a good idea to test various update sceneries, to make sure things work as expected (for example, if you already have M6 installed, then once M7 is ready, test to be sure it updates to M7 as expected.

Test staging as a pseudo composite

After confirming staging repository is correct, its often useful to then also (re)enable the released repository. Then, on install dialog, you can select "all available sites" and this effectively simulates what the eventual, final composite repository will look like to end users. There are cases, especially during initial development, where invalid features will show up. For example, if a feature was removed, renamed, its version reduced, or its category changed from one milestone to the next, it might still show up in the composite, and that might interfere with correct installation (see bug 314165), if not merely be confusing to end users. If there is a serious problem due to the composite, please open a bug in cross-project component and we'll decide if the composite should be changed or reduced to allow for correct installation.


Test EPP Package updates

For a release, EPP packages are added to the common repository via composite, so its exact (final) location is transparent to end-users.

But before a release, similar to above, you can use a pseudo composite to test if an EPP Package updates as expected. You'd need to follow (or ask on) the epp-dev list to know details, but in addition to the above staging repo, you would add (and enable) EPP's staging repo, which would usually have a form similar to the following:

 http://download.eclipse.org/technology/epp/packages/indigo/SR1.281/

Why do we use composites anyway, if there are potential problems with them?

Two reasons. While the eventual, initial released repository, in June, will definitely consist of just one set of released, consistent features, later in September, and then February, we use composites to add maintenance, so testing composites early is a good idea to make sure there are no bugs in p2, etc., that need to be fixed.

Secondly, someone may be using a previous milestone as a runtime target, and once we remove it, it will invalidate that target, so we do not want to force everyone to "move up" all at the the same time. Some developers may need a few weeks to transition to the latest code. To save space, we do not try and save all previous milestones, though. Our goal is to maintain about 3 milestones in the composite. For example, if we had M4, M5, and M6 in the composite, once M7 came out, we'd only have M5, M6, and M7. But the number can not be guaranteed. If there are serious problem with providing the composite we will reduce it to just two, or even to just one, to make sure the repository is usable for testing and continued development.

Once I update my .aggrcon file, how can I start a build?

You don't need to. The build will start automatically, once you check in a .aggrcon file. The build project is checked every 15 minutes to see if any changes, and if so an aggregation build will start. It takes approximately 2 hours to run.

But what if I really want to kick off the build myself?

If you can't wait 15 minutes, you can start the build your self. Anyone that has authorization to check in a build file, should have authority to manually start a build.

Plus, there are some cases where someone may need to kick off a build manually. For example, if a build fails due to network issues. Another common case is that a build may fail, even though the contribution file is correct, the repository it points to might have had an error. Once the repository is corrected, there's no automatic mechanism to detect that change, so after the repository is corrected, a new build has to be manually started (that, or the contribution file "touched" and then checked in again). To manually start a build, just click the "Schedule a Build" button at the build status control page.

You need to login (with your committer ID):

SimRel login.png


Then click the "Schedule a Build" buttton:


SimRel build.png

A build failed message says it can not find xyz.feature.group, but I have nothing with "feature.group" in the name?

The suffix ".feature.group" is added to feature names, to refer to the whole feature ... the feature files themselves, but also all its included bundles and features ... to help distinguish it from the literal feature files. So "xyz.feature.group" just means the "xyz feature", conceptually. See Eclipse Help for detailed information about metadata.

A build failed message says it can not find version 1.2.3.v9 but I can see 1.2.3.v9 on the file system?

The key file, the one that "drives" P2, is the content.jar/xml file. Be sure to check the version numbers there. If, inside it, the installable unit (often a feature) says version="1.2.3.v8", then P2 will look no further and conclude that the 'v9' feature it is looking for is not there. This is usually a sign your meta data needs to be re-generated to match the contents on the file system.

How is a final build made "invisible" until release?

Web download pages?

You can put the zips in their download directory, so those large items can replicate to mirrors, but don't use any HTML that would cause them to be displayed as links to an end user. True, if someone knew the exact URL they could still get it, but the idea is that the URL is not widely announced or visible, so even if a few download it, it is not hit by thousands of downloads. Then, on release day, you'd update the HTML pages to make the downloads visible to browsers and click-able by users.

Tip : you can hide some particular builds from download pages using the hidden.txt file in downloads directory ( see for example emft : http://dev.eclipse.org/viewcvs/viewvc.cgi/www/modeling/emft/downloads/hidden.txt?root=Eclipse_Website&view=log )

P2 repositories?

There is a few ways to accomplish this, depending on if you have composite or simple repositories, but they all involve promoting the "main" parts of the repository (the artifacts, usually "plugins" and "features" directory) to their final location so those meaty parts can be replicating to mirrors, but do not put the metadata (usually content.jar and artifacts.jar) to their final location until "release day". p2 doesn't "see" the artifacts, until it can read the metadata.


I see that xyz is being pulled into the repository, how can I see who is pulling it in?

There are times when someone may see an old, or inappropriate, bundle being "pulled in" to the aggregation build, and they want to find out how or why it is being included. While the full rules for how p2 decides something is required, or satisfies a requirement, is beyond the scope of this FAQ, there are two ways to get started narrowing it down. Just to have a concrete example, I will use the use-case provided by bug 357171 to show a specific case, where some old (unsigned) version of org.eclipse.emf.teneo.hibernate.libraries was being included.

The quick approximate check of where its coming from

Check the console log for a message about mirroring the problematic bundle, and then look back "up" in the log, to see what repository is being processed at that point.

For, our example, searching for 'org.eclipse.emf.teneo.hibernate.libraries' we find the following line:

    [exec] - mirroring artifact osgi.bundle,org.eclipse.emf.teneo.hibernate.libraries,1.0.1.v200907090915 

Then, looking back "up" in the log, we see the following repositories were being used to pull things from:

    [exec] Mirroring meta-data from from file:///home/data/httpd/download.eclipse.org/modeling/emf/updates/interim
    [exec] - mirroring meta-data reference http://download.eclipse.org/modeling/mdt/updates/
    [exec] - mirroring artifacts reference http://download.eclipse.org/modeling/mdt/updates/
    [exec] - mirroring meta-data reference http://download.eclipse.org/modeling/emf/updates/
    [exec] - mirroring artifacts reference http://download.eclipse.org/modeling/emf/updates/
    [exec] - mirroring meta-data reference http://download.eclipse.org/modeling/emft/updates/
    [exec] - mirroring artifacts reference http://download.eclipse.org/modeling/emft/updates/
    [exec] Mirroring artifacts from from file:///home/data/httpd/download.eclipse.org/modeling/emf/updates/interim

In some cases, this might be sufficient to track down the problematic contributor.

The detailed, exact check on requirements

If the log doesn't help, the definitive source of "requirements" can be seen in the content.jar/xml. It can be tricky to "find" the content.jar/xml file, depending on composites repositories, if compressed or not, etc., but for our example, from a temporary maintenance repository, you can download it from

 http://download.eclipse.org/releases/maintenance/content.jar

Once unzipped, you can search for your bundle, in our case. searching for 'org.eclipse.emf.teneo.hibernate.libraries'.

You'll certainly find the IU that defines the bundle

 <unit id='org.eclipse.emf.teneo.hibernate.libraries' version='1.0.1.v200907090915' singleton='false'>

That IU, by itself, doesn't help narrow down who is requiring it. But, continue to search.

The easy case of "Require-Bundle"

If you find a match in a "requires" element, you can see what bundle "requires" your bundle. If found, those matches would start with "<required" element and include the name of the bundle. In these cases, there is some bundle that uses "Require-Bundle" and should be easy to resolve with who ever is doing that.

The harder case of "Import-Package"

If you find no direct "<required" match, then that means p2 is deciding your bundle "fits" the requirements due to the packages it provides (exports), and some other bundle (or bundles) "imports" those packages. If there are just a few, uniquely named packages provided by your mystery bundle, you can search the content.xml file for who "requires" those packages and figure out what to do from there.

The hardest case

But, our example does not have such a happy ending. The IU 'org.eclipse.emf.teneo.hibernate.libraries' provides 264 packages! Some of those are fairly distinctly named, but some are very common, and very commonly "imported" by other bundles. While I did not try, it would be hard to narrow down to just a few other bundles that might be requiring the many exported packages. I do not know the exact details, but p2 does have some heuristics where it will give extra weight to include a bundle that satisfies the most requirements -- so, any bundle that provides (exports) a huge number of common packages is in danger of getting included unintentionally by p2.

Back to the quick approximate findings

Another approach is to go back to those repositories implied in the "The quick approximate check" above.

Ideally, it would be possible to provide a more specific repository for the CBI aggregator so it only looks in more currently correct repositories. For example, (and, I emphasize 'example', since I am unsure of Modeling project's policies and procedures) a repository for "../emf/updates/indigo/" might be more appropriate for indigo contributions than a high level "../emf/updates/" repository. Or, as another example, perhaps "../emf/updates/interim" repository is actually incorrect for Indigo contributions. In that case, you could search the *.aggrcon files to see who is using or providing that repository and work out a better alternative.

A caution on how not to solve

For cases such as this, where some old, inappropriate bundle is being included, one solution, one might think, would be to surgically remove it from its original repository. This is almost never a good idea, since someone, somewhere, might be depending on it being there ... released repositories are forever, or, at least, are supposed to be ... so great care is needed in changing released repositories.


What if we need a rebuild, past our deadline?

Everyone knows plans are just plans, and occasionally exceptions are required. While we can't cover every exception (after all ... they are exceptions) there are the some general principles of deciding if an exception is required. The same ideas apply to milestones, release candidates, and service releases, but the impact levels are higher the further along a cycle is. If possible, naturally, it is best to handle problems within your own project, say, by providing a patch feature or update to your own project's software repository, but in some cases, that's not possible ... this FAQ entry is about those cases.

Questions to ask yourself (and answer, in your request for an exception or rebuild)

Not necessarily an exhaustive list ... but, some common questions:

  • Is the bug something that effects IP or similar "legal requirements" from the Eclipse Foundation?
  • Is the bug something that prevents other projects from working correctly?
  • Is the bug something that causes install, or update, to fail, or otherwise leave an installation in a bad, unfixable state?
  • Is the bug something that can not be solved by a patch feature, applied by users or adopters after the release?
  • Is the bug a regression, from previous release?
  • Do other projects have to recompile, once a bug is fixed? For example, are constants changed? APIs? version numbers?
  • Do other projects need to retest, once a bug is fixed? That is, is it something that could potentially effect others ... such as a change in timing or synchronization of some "notification" to listeners? Or something clearly "internal" to your own code?
  • Do you have the support, approval, or review of your Project Lead and PMC? (Or, otherwise follow what ever your project's rampdown process is?)
  • Does it affect an EPP Package? Or just the common repository?

Past our +n day, but before window closes?

If we are still within the "drop window" for a deadline, but you are past your particular +n day, simply post a note to cross-project list, with any relevant questions and answers from above list, the bug number, and then just do it. (No need for further approval or coordination.)

Keep in mind that spinning new builds past your deadline can result in a lot of work for downstream projects and consumers as they make last minute adjustment to your change. Projects are generally willing to accommodate these changes as much as they can, but please keep this in mind and only do it when absolutely necessary.

Past the drop window?

In this case, it is completely past the drop window, after EPP packages have been built. In this case, you still need to post to cross-project list, with bug number, and relevant questions and answers from above list, but now explicit review/permission from Planning Council is also required. Please follow the Planning Council Exception Process, but in cases of "tight timing", the Planning Council has authorized the Planning Council Chair (currently David Williams) to make the initial decision and allow others to review later or in parallel.

Common errors and what to do about them

There are some errors that occur during aggregation builds that are pretty common and asked about frequently so will list some here in this FAQ. Please add others, if you notice any missing, and/or improve the answers if any of these are incomplete or not helpful enough.

Aggregation model is inconsistent

The title is typically how the error message always begins, and then has several lines of hard to read notations, which are probably easy to read if you know EMF real well. :) There are two common reasons for this error, both are related to the fact that in some cases, the *.aggrcon file and the simrel.aggr file must *both* be updated.

An interesting twist of this problem is that when it occurs there is no "blame mail" sent out. The reasoning is that "since the model is bad it is hard to trust the email data from the model". Another good reason to use Gerrit, since then you know (pretty much) the error is yours, if this error occurs after your contribution. Otherwise, the overall Simultaneous Release Engineer must figure out the offending party -- which is pretty easy to do if they use the CBI Aggregator Editor.

Solution:

If the hard-to-read notations mention something about "missing proxy" then it is likely a "contact email" (or a feature id) was added to the projects aggrcon file, but was not added to the simrel.aggr file. It is easiest to use the CBI Aggregation editor (on the simrel.aggr file) to add/remove the feature or the contact and also add to a contribution or category.
If the hard-to-read notations mention something about custom categories and features that "do not refer to each other" then similar to the above, someone tried to add a feature to a category, but changed only their aggrcon file and not the simrel.aggr file. The solution is the same, use the Editor!

In both cases above, after the change is made, but the simrel.aggr file and the project's aggrcon file should show "changed" and both should be committed together.


Invalid pack.gz file

The actual error message is not nearly so clear as the title of this section, but more often says something like the following (from bug 492904)

org.eclipse.core.runtime.CoreException: Unable to unpack artifact osgi.bundle,org.apache.hadoop.zookeeper,3.3.3.v201105210832 in repository file:/home/hudson/genie.simrel/.hudson/jobs/simrel.neon.runaggregator.BUILD__CLEAN/workspace/aggregation/final/aggregate: Invalid content:org/apache/zookeeper/server/upgrade/UpgradeMain.class
Caused by: org.eclipse.osgi.signedcontent.InvalidContentException: The file "org/apache/zookeeper/server/upgrade/UpgradeMain.class" in the jar "/home/hudson/genie.simrel/.hudson/jobs/simrel.neon.runaggregator.BUILD__CLEAN/workspace/tmp/signatureFile6913692817837489877.jar" has been tampered!

The problem is that the "pack.gz" file is invalid. For this example the bundle is probably something like "org.apache.zookeeper.server", but technically the error only mentions the package name, and the "temp location" of the jar file.

An interesting twist to this problem is that p2 itself will not complain if someone is simply "installing" the bundle, since if an error occurs with the "pack.gz" version of a bundle, p2 automatically falls back and uses the "jar" version of a bundle. The reason that it is considered a blocking error during aggregation is that most large repositories should be "perfectly correct". It is no big deal if one out of several thousand was wrong, but imagine if 25% of those thousands were "invalid pack.gz files". The end-user will end up wasting a lot of time doing updates, and the infrastructure bandwidth would be increased for no good reason.

Solution:

There are 2 or 3 or 4 possible solutions.

The jar may may be being "re-signed" during the project's build. Re-signing actually does work sometimes but not when "pack.gz" is involved. Hence, one solution is for the project to stop re-signing the bundle (which is difficult for some build system, but which is now automatic for most cases at Eclipse.org).
Pack200 (that produces the pack.gz file) does not work for each and every possible piece of Java byte codes. It should, but it doesn't. So a few solutions here:
The best (or easiest) is for the project to specify at build-time not to use pack200 with that particular bundle.
Even better (but very difficult, and only worth if it is a very large bundle), is that pack200 has options to specify to do the "packing" slightly differently. For this example, as one use of the options, it might work to tell pack200 not to pack the "org/apache/zookeeper/server/upgrade/UpgradeMain.class". (BTW, I do not even know if that is possible with Tycho. :)
Another "quick fix" solution is to run a post build step to remove that specific pack.gz file from the project's repository. This is not as simple as deleting the pack.gz file, since it is specified in the metadata files as "being available" but there are some simple p2 ant tasks that can be used to remove it correctly (as an example, see the example mentioned in bug 492904.

Additional Information

Back to the top