| 1 |
[[!meta title="DEP-10: parallelized ('rolling') release management"]]
|
| 2 |
|
| 3 |
Title: parallelized ('rolling') release management"
|
| 4 |
DEP: 10
|
| 5 |
State: DRAFT
|
| 6 |
Date: 2011-04-30
|
| 7 |
Drivers: Sean Finney <seanius@debian.org>,
|
| 8 |
Raphaël Hertzog <hertzog@debian.org>
|
| 9 |
URL: http://dep.debian.net/deps/dep10
|
| 10 |
License: GPL
|
| 11 |
Abstract:
|
| 12 |
Proposal for changes to release management methodology and infrastructure
|
| 13 |
to remove the requirement that the testing suite must freeze during
|
| 14 |
the release process.
|
| 15 |
|
| 16 |
<a name="introduction">
|
| 17 |
# Introduction / Problem scope
|
| 18 |
|
| 19 |
Currently, as the project nears a new stable release, a freeze is instituted
|
| 20 |
on the testing suite. The freeze is put in place to allow the release team
|
| 21 |
to focus on resolving the remaining Release Critical (RC) bugs for the next
|
| 22 |
stable release, and at the same time to prevent regressions from new uploads.
|
| 23 |
Typically the freeze begins as an advisory "soft freeze", which over
|
| 24 |
time increases in strictness and levels of enforcement.
|
| 25 |
|
| 26 |
Unsuprisingly, as the strictness of the freeze increases, there is an
|
| 27 |
inversely proportional decrease in other non-release targeted maintainer
|
| 28 |
activity. Since unstable still is the preferred route for packages to
|
| 29 |
reach the new release during this period, maintainers are highly discouraged
|
| 30 |
and in some cases prevented from doing non-release targeted activities
|
| 31 |
in unstable.
|
| 32 |
|
| 33 |
The reduction of such non-release activity is viewed as problematic in
|
| 34 |
this DEP, for some inter-related reasons:
|
| 35 |
|
| 36 |
* New features and innovations are put on hold, or at least not commonly
|
| 37 |
available, until after the release is made.
|
| 38 |
* Overall Maintainer activity decreases as freezes persist.
|
| 39 |
* Potential userbase is lost (missing feature X, switch distro).
|
| 40 |
* The volatility of testing/unstable increases (and thus quality decreases)
|
| 41 |
with a deluge of new uploads after the freeze is lifted.
|
| 42 |
|
| 43 |
As Debian is well known for taking a "release when it's ready" approach, the
|
| 44 |
freeze periods are generally known to last considerable amounts of time.
|
| 45 |
Consider the last three freezes:
|
| 46 |
|
| 47 |
* squeeze: 4 months
|
| 48 |
* lenny: 7 months
|
| 49 |
* squeeze: 6 months
|
| 50 |
|
| 51 |
This means, in rough terms, that the when testing thaws, that the Debian
|
| 52 |
project may be starting from a state half a year behind comparable
|
| 53 |
distributions.
|
| 54 |
|
| 55 |
While the resulting stable releases are well known for their high level
|
| 56 |
of quality and stability, it is at a considerable cost. The project is,
|
| 57 |
in essence, starting from a 6 month standstill, will be similarly
|
| 58 |
out of date with comparable distributions. Furthermore, both the standstill
|
| 59 |
as well as the subsequent rush of uploads will introduce further delays
|
| 60 |
to the next release, as release goals will likely be set relative to
|
| 61 |
the starting point.
|
| 62 |
|
| 63 |
# Past and present release methods
|
| 64 |
|
| 65 |
## Frozen (< 2000)
|
| 66 |
|
| 67 |
Before the introduction of testing, Debian used a simple release process
|
| 68 |
where the unstable suite was snapshotted into a `frozen` suite. This suite
|
| 69 |
would then be used exclusively for preparing the next stable release, with
|
| 70 |
unstable continuing in parallel.
|
| 71 |
|
| 72 |
Before freeze Freeze Release
|
| 73 |
|
| 74 |
[unstable/sid]--------------------------------------------------------------
|
| 75 |
\
|
| 76 |
[frozen].-.-.-.-.[stable/R_N].-.-.-.-.-.-.-.-.-.-.-.-.-.
|
| 77 |
|
| 78 |
[stable/R_N-1].-.-.-.-.-.-.-.-.-.-.-.[oldstable/R_N-1].-.-.-(EOL)
|
| 79 |
|
| 80 |
--------: Normal activity. Standard rules for uploads and migrations.
|
| 81 |
.-.-.-.-: Release targeted activity. Freezes and limited uploads.
|
| 82 |
\ \ \ \ : Package migration activity.
|
| 83 |
|
| 84 |
### Use cases with `frozen`
|
| 85 |
|
| 86 |
Users were divided into two sets, those using `unstable` and those using
|
| 87 |
`stable`. Very few users would use both, as the two lines of development
|
| 88 |
would quickly diverge from each other.
|
| 89 |
|
| 90 |
### Benefits with `frozen`
|
| 91 |
|
| 92 |
* Work in unstable continued, unaffected by the freeze.
|
| 93 |
|
| 94 |
### Problems with `frozen`
|
| 95 |
|
| 96 |
* Release work started from a buggy suite.
|
| 97 |
* Duplicated effort in unstable/frozen.
|
| 98 |
* RM had more work/responsibilities, and was prone to burn-out.
|
| 99 |
|
| 100 |
## Testing (2000-Present)
|
| 101 |
|
| 102 |
The testing suite was introduced in Debian between the release of potato
|
| 103 |
and woody, in the fall of 2000[[1]]. The goal was to provide
|
| 104 |
a suite that was in a better state for release preparation, by having
|
| 105 |
both automated and manual tools to keep down the level of bugs and
|
| 106 |
general volatility.
|
| 107 |
|
| 108 |
As a pleasant and convenient side-effect, the new suite also provided
|
| 109 |
a "slightly less buggy unstable" for developers and end-users, who wanted
|
| 110 |
newer software/features not available in stable, but wanted some level
|
| 111 |
of protection to the relatively unpredictable nature of unstable.
|
| 112 |
|
| 113 |
Before release Freeze Release
|
| 114 |
|
| 115 |
[unstable/sid]----------.--.--.--.-.-.-.-.-.-.------------------------------
|
| 116 |
\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
|
| 117 |
[testing/R_N]----------.-.-.-.-.-.-.-.-.-.-.-.[testing/R_N+1]---------------
|
| 118 |
/ / \
|
| 119 |
/ / [stable/R_N].-.-.-.-.-.-.-.-.-.-.-
|
| 120 |
/ / / / /
|
| 121 |
[R_N p-u].-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
|
| 122 |
|
| 123 |
[stable/R_N-1].-.-.-.-.-.-.-.-.-.-.-.-.-.-[oldstable/R_N-1].-.-.-(EOL)
|
| 124 |
|
| 125 |
--------: Normal activity. Standard rules for uploads and migrations.
|
| 126 |
.-.-.-.-: Release targeted activity. Freezes and limited uploads.
|
| 127 |
\ \ \ \ : Package migration activity. Spacing of marks is a rough
|
| 128 |
indication of frequency.
|
| 129 |
|
| 130 |
During the freeze, the testing suite becomes entirely dedicated to the
|
| 131 |
release work. In practice, this also means that unstable is also to
|
| 132 |
some extent frozen from non-release activity for many packages, since
|
| 133 |
it still serves as the main route to testing for new uploads.
|
| 134 |
|
| 135 |
### Use cases with `testing`
|
| 136 |
|
| 137 |
The introduction of the new suite also introduced a type of continuum
|
| 138 |
in which users now have more flexibility in selecting what to have installed.
|
| 139 |
|
| 140 |
* `unstable`: only the "bleeding edge" of Debian updates.
|
| 141 |
* `testing`: only the "leading edge" of Debian updates.
|
| 142 |
* `testing/unstable`: A hybrid solution of the two previous use cases,
|
| 143 |
allowing for a reasonable balance between them.
|
| 144 |
* `stable`: The latest stable release.
|
| 145 |
* `stable/testing`: the stable release with perhaps some newer packages
|
| 146 |
installed, typically also making use of "APT pins" to prevent unwanted
|
| 147 |
upgrades. Less common now with the inclusion of backports and volatile
|
| 148 |
as official Debian services.
|
| 149 |
|
| 150 |
### Benefits with `testing`
|
| 151 |
|
| 152 |
* Release branch has far fewer bugs at the start of the release process.
|
| 153 |
* (Ideally) shorter release process with fewer problems.
|
| 154 |
* Large user-base for testing stable-targeted fixes.
|
| 155 |
|
| 156 |
### Problems with `testing`
|
| 157 |
|
| 158 |
(See [Introduction](#introduction))
|
| 159 |
|
| 160 |
[1]: http://lists.debian.org/debian-devel/2000/08/msg00906.html
|
Parent Directory
| 
Revision Log