| Summary: | Regression: p2 downloads are much slower due to picking remote artifacts even when a local file: URL is available | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Product: | [Eclipse Project] Equinox | Reporter: | Meng Xin Zhu <kane.zhu> | ||||||
| Component: | p2 | Assignee: | DJ Houghton <dj.houghton> | ||||||
| Status: | RESOLVED FIXED | QA Contact: | |||||||
| Severity: | critical | ||||||||
| Priority: | P3 | CC: | helmut.haigermoser, jeffmcaffer, kane.mx, mober.at+eclipse, pascal | ||||||
| Version: | 3.6.1 | Keywords: | performance | ||||||
| Target Milestone: | 3.7 M6 | ||||||||
| Hardware: | PC | ||||||||
| OS: | All | ||||||||
| Whiteboard: | |||||||||
| Bug Depends on: | |||||||||
| Bug Blocks: | 335610, 408208 | ||||||||
| Attachments: |
|
||||||||
|
Description
Meng Xin Zhu
Ouch, if this is true than 3.6 is severely broken and must be fixed right away, favoring "http" over "file" repos will kill performance! Can we get a reply to this ASAP, and bump it to highest priority if your analysis confirms the issue? Helmut Please provide a test case showing the problem and a patch. Created attachment 187497 [details]
test case and proposed fix
Let bundle p2.test require the packages related to osgi http service, and add http.port vm argument in the launch.
Let me know what you think. Thanks.
Created attachment 187498 [details]
test data
unzip it, then put the folder under 'testData'
CQ:WIND00252176 From what I understand, this is a severe regression in Eclipse 3.6, compared to Eclipse 3.5. The regression affects products that install from both a local repo (like a shipped DVD) and an online repo (like an ESD / update pack site). Effect of the regression is that the install performance can be severely reduced by downloading artifacts that already exist locally. Customers likely wouldn't accept such a regression, so I'm setting "critical" severity. I've also updated the summary to show the end-user effect, previous summary was: DownloadManager.start() doesn't sort the repositories based on the predefined priorities when trying to download the artifacts from the repositories one by one Please consider the testcase and fix that Meng Xin has provided. Released a modified patch to HEAD. Thanks for the investigation. Closing. Thanks so much, DJ! Could a backport to 3.6.2 be considered? We are very close to releasing a product (based on 3.6.x), and this issue is a showstopper for us. For sure we'll have to patch the Platform of our RCP installer ourselves if this doesn't get addressed in mainstream. Sounds good to me. I've created bug 335610 for this. I will attach a patch and then ask a PMC member for a +1. |