Bugzilla – Full Text Bug Listing
| Summary: | Target Platform Resolution often fails for update sites at eclipse.org | ||
|---|---|---|---|
| Product: | Community | Reporter: | Maximilian Koegel <mkoegel> |
| Component: | Servers | Assignee: | Eclipse Webmaster <webmaster> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | major | ||
| Priority: | P3 | CC: | cletavernier, david_williams, denis.roy, jfaltermeier, mkoegel, stepper, torkildr |
| Version: | unspecified | ||
| Target Milestone: | --- | ||
| Hardware: | All | ||
| OS: | All | ||
| Whiteboard: | |||
|
Description
Maximilian Koegel
In addition to Nick's good advice to Arch. Council list, here are some other comments: 1. when a problem is seen, be sure to try something similar to wget http://download.eclipse.org/releases/oxygen/201706281000/content.xml.xz which should tell you if the main initial request is being re-directed. Even here in North Carolina as of right now (08/10/2017, 20:50 Eastern), that request replies with "302 Moved Temporarily" and then is being redirected to http://mirror.math.princeton.edu/pub/eclipse//releases/oxygen/201706281000/content.xml.xz [Actually, each time the request is sent, I have just now seen that is redirects to a different mirror.] I've never seen that initial redirect strategy work well. I *think* it *might* be because then "mirror.math.princeton.edu" (or whatever) becomes the "fallback URL" which assumes it is as perfect as "download.eclipse.org". (Half guessing the reason why it does not work well, but 100% sure I have never seen it work well. Perhaps a p2 committer knows more if it should or what it assumes?) 2. for more p2 options to enable more debug and trace output to see which mirrors are being used during a p2 operation and which, if any, are causing problems see https://wiki.eclipse.org/Equinox/p2/p2.mirrorsURL#How_to_test.3F 2A. There is a possibility of a "bad mirror" that throws everything off. (For example, one that has partial, but not full content -- I don't think p2's heuristics cover that case well -- and other cases such as sending a bundle named "abc.php.bundle.jar" as "php" content, instead of as binary content). 2B. I didn't see mention of which repositories were causing problems, but, in the past, if the p2.mirrorsURL was set incorrectly in a repository, it causes a lot of churning and everything then tries the default fallback URL. 3. The "Foundation" (SimRel engineer) really should clean up the old directories from .../releases/oxygen. Currently 11 directories there are taking approx. 28G of space. Only one is needed (201706281000) taking about 2.4G. Some mirrors might be running low on space. Just a wild guess. (I suspect in 99% of cases a measly 20G or so would not matter to a mirror-class server -- but, I do not know for sure). 4. I'm sure most readers know about setting -Declipse.p2.mirrors=false which I mention not so much as a long-term solution, but those having issues should be mentioned in this bug report if that setting solves the problem or if the problem is the same. HTH, Good Luck, For future reference, Nick's advice to Arch. Council list is at the following URL: https://dev.eclipse.org/mhonarc/lists/eclipse.org-architecture-council/msg03482.html For target platforms, Mickaël's tool for Target Platform Definition [1] includes a "Set as target" context menu which is much more efficient and reliable than Eclipse's mechanism. So, whenever I have the opportunity, I use *.tpd files instead of *.target (*.target will be generated), and use the context menu to set/switch TPs. One of the differences is that the TPD Tool uses a local cache for bundles, but that's not the only difference. I suspect the algorithm to actually resolve and set the TP is also simpler (Thus faster and almost 100% reliable), although I've not looked into it. I don't know how much of this could be ported directly to the Eclipse Target Platform framework, but it might be worth investigating further [1] https://github.com/mbarbero/fr.obeo.releng.targetplatform As bug 517294 is resolved, you will not see this anymore. |