Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [p2-dev] Product publishing and product update

Hello again,

 

I am continuing with some experiments along the directions that Jeff gave me. I encountered several problems for which I cannot find an explanation. For example, I tried to update the product after incrementing its version in the repository. The update failed again because it lists among its requirements a tooling configuration unit which is a singleton. It looks quite simple:

    <unit id='tooling<product name>.configuration' version='<product version>'>

      <provides size='1'>

        <provided namespace='org.eclipse.equinox.p2.iu' name='tooling<product name>.configuration' version='<product version>'/>

      </provides>

      <touchpoint id='null' version='0.0.0'/>

    </unit>

 

Note that this is generated by the product publisher and cannot be avoided. I don’t have any idea what the purpose of such a basic unit could be but being a singleton and a requirement of the product, it stops the update of the whole product because there is already an IU installed with the same name on the system (actual message from p2 director says “Only one of the following can be installed at once”, concerning this IU).

 

Can anybody tell me why is this configuration unit created at all on publishing ?

 

 

In general, I am very surprised to see how many problems I encounter to implement a “simple” product update given the fact that p2 supports updates of features and bundles out of the box. So far, the most direct approaches I tried failed completely:

-          If I try to update, preserving the same product version (as it is fixed in the .product descriptor), it fails because of conflicting versions of the requirements.

-          If I try to update with an increased version of the product, then the singleton configuration unit stops me.

 

So it seems that my initial concept how the product update should be done is wrong. But then how new versions of products are supposed to be shipped to customers to be consumed immediately by p2 ? How are the customers supposed to perform updates of the whole product (not by individual bundles and features) ?

 

 

Best regards,

Shenny

 

 

 

From: p2-dev-bounces@xxxxxxxxxxx [mailto:p2-dev-bounces@xxxxxxxxxxx] On Behalf Of Jeff McAffer
Sent: Friday, September 24, 2010 4:36 AM
To: P2 developer discussions
Subject: Re: [p2-dev] Product publishing and product update

 

There are a couple sides to this.  One is that if you have Product X v 1.2.3.20100923, that should mean something. If you allow ranges as described, then two users installing X 1.2.3.20100923 may not get the same actual software installed. Variation is introduced for example, if user 1 has access to a different set of repos than user 2 or there is a network error for user 1 but not user 2 or the single repo changed between when user 1 and user 2 did their install.

 

Of course, these behaviours *could* also be exactly what you want but certainly some folks free at this non-determinism as a support nightmare.

 

Anyway, looking at features, they allow for things to be *included* or *required*.  Included things have exact version ranges while required things have, generally, wider ranges. Traditionally the notion was that on install, the things *included* by the feature were installed whereas the things *required* merely had to be there. Early update manager didn't even help you find/get/install the required things.  That was goofy so we provided a means for users to say "yeah, get the required stuff also".  Now with p2 we do this automatically without involving the user.  So much for context...

 

It would be reasonable to allow ranges on product content but it would also force the product designer to be very aware of the consequences pointed out at the beginning of this message. I honestly don't know what people would do naturally or what guidance we could/should give them (e.g., what's the default?).

 

Back to your original topic, there is also the possibility of producing new versions of your product that identify the new versions of the components. Product production and distribution in p2 is very light weight and users would see this as incoming new versions of the product (that they know about) vs changes to random components (that they may well not even know exist).  What would you say as the user of some banking product if told that there was a new version of EMF?  "WFT?!"

 

Scenarios vary.  If that does not work for you, you can insulate your product by making it consist of one feature. In that feature, *require* everything that you want to be updatable, include the stuff you want to be fixed (or put this stuff directly in the product).  The product will be bound to the one version of your container feature and the container feature can use ranges.  Beware the problems outlined above with non-determinism.  Note that you can also usethe p2.inf file to do this.  Andew Niefer did a couple blog posts on this a while ago

 

Good luck

Jeff

 

 

On 2010-09-23, at 12:13 PM, Yousouf, Shenol wrote:



Hi all,

 

I noticed that product publishing always sets requirements for a fixed version of the contained bundles/features, i.e. the defined range has its lower and upper boundaries equal like this:

<required namespace="org.eclipse.equinox.p2.iu" name="TestBundle"range="[1.0.0.201009171510,1.0.0.201009171510]" />

while I need something like this:

<required namespace="org.eclipse.equinox.p2.iu" name="TestBundle"range="[1.0.0.201009171510,2.0.0)" />

or even this:

<required namespace="org.eclipse.equinox.p2.iu" name="TestBundle" range="1.0.0.201009171510" /> (which means “any version > 1.0.0.201009171510)

 

 

The .product file format does not support a way to specify a range for its components, only an attribute for a fixed version. The product publisher also has no notion how to generate version ranges – it simply sets the range boundaries equal to the component version (see method AbstractPublisherAction.createIURequirements() for reference). So far, I cannot find a way how to workaround this issue and in my opinion it as a limitation of the product definition concept.

 

Why is this so important ? The use case is like this:

I am developing a product consisting of several components which is getting published on an update site on a regular basis. The components receive frequent updates in the p2 repository and their versions are incremented which is reflected in the requirements of the published product. However, once I install this product, I cannot apply updates to the system any more. The updates are refused because version ranges of the requirements for the installed and the updated products do not intersect which seems to make them incompatible.

 

This wouldn’t be the case if it was possible to define open ranges in the product file. For example, the installed product would require a specific component in version range [1.0.0, 2.0.0) while its new version would require it in the range [1.1.0, 2.0.0). This would allow the update to pass because obviously range [1.1.0, 2.0.0) is compatible with (falls into) range [1.0.0, 2.0.0). The way they are generated now is [1.0.0, 1.0.0] for the old product and [1.1.0,1.1.0] for the new one. Since these two ranges do not intersect, the update is not possible.

 

In short, I have two issues and hope to receive some advice from you how to address them:

  • Is it possible to define a product with extended version ranges of its components ?
  • What makes product versions compatible for update ? Why changed version requirements, which come as a natural result of the publishing process, do not allow the product to get updated to the higher version of its included components ?

 

 

Best regards,

Shenny

 

_______________________________________________
p2-dev mailing list
p2-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/p2-dev

 


Back to the top