Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 1 | This wiki page is maintained by the Go team. |
| 2 | Please [send comments to golang-dev](TODO://LINK) |
| 3 | or [file issues](https://golang.org/issue) instead of making changes directly. |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 4 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 5 | Short link: https://golang.org/s/release. |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 6 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 7 | ## Overview |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 8 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 9 | After fourteen months elapsed between Go 1.0 and Go 1.1, the Go team |
| 10 | adopted a release schedule to streamline the process of working on, |
| 11 | finishing, and issuing releases. |
| 12 | The overall goal is to make a major release every six months, which breaks down into |
| 13 | three months of general development followed by three months of testing and polishing |
| 14 | known as the release freeze. |
| 15 | A release is maintained by issuing minor releases to correct |
| 16 | critical problems such as crashes or security issues. |
Andrew Gerrand | 847452d | 2015-06-25 22:30:32 +1000 | [diff] [blame] | 17 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 18 | Note that this page documents what we intend to do for upcoming releases. |
| 19 | If you are familiar with what we have done in the past, |
| 20 | pay particular attention to the differences described in the |
| 21 | [history section](#history). |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 22 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 23 | ## Timeline |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 24 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 25 | The current release cycle is aligned to start on February 1 and August 1 of each year. |
| 26 | The target milestones for a release cycle are as described below. |
| 27 | We try to hit the targets as closely as possible, while still delivering |
| 28 | a quality release. |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 29 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 30 | Betas, release candidates, and releases have traditionally been cut and issued |
| 31 | mid-week, often on Wednesdays. |
| 32 | We avoid Mondays to give the Go team a work day before to prepare for the cut, |
| 33 | and we avoid Fridays to give the Go team a work day after in case the cut runs |
| 34 | into unexpected problems. |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 35 | |
Russ Cox | 76b4923 | 2016-01-28 10:20:05 -0500 | [diff] [blame] | 36 | #### January 15 / July 15: Planning for release begins. |
Caleb Spare | 5e9e0b2 | 2015-06-24 23:57:39 -0700 | [diff] [blame] | 37 | |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 38 | Planning of major work for upcoming release cycle is announced |
| 39 | on [golang-dev](https://groups.google.com/group/golang-dev). |
| 40 | |
| 41 | Examples: |
| 42 | [Go 1.4](https://groups.google.com/d/msg/golang-dev/eqBihsj7x-k/3h59pc8FDAoJ), |
| 43 | [Go 1.5](https://groups.google.com/d/msg/golang-dev/2ZUi792oztM/GNsw1i76LhsJ), |
| 44 | [Go 1.6](https://groups.google.com/d/msg/golang-dev/vNboccLL95c/XOD3vHKOCQAJ). |
| 45 | |
Russ Cox | 76b4923 | 2016-01-28 10:20:05 -0500 | [diff] [blame] | 46 | #### February 1 / August 1: Release work begins. |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 47 | |
| 48 | This milestone is delayed if the previous release has not yet been issued, |
| 49 | but note that later milestones are not delayed. |
| 50 | That is, failure to get one release out on time takes time away from |
| 51 | the development cycle of the subsequent release. |
| 52 | |
| 53 | Note that incoming bug reports should be handled and bugs fixed during |
| 54 | ordinary release work. |
| 55 | It is not appropriate to leave all bug fixes for the release freeze. |
| 56 | See the [history](#history) section for more discussion. |
| 57 | |
Russ Cox | 76b4923 | 2016-01-28 10:20:05 -0500 | [diff] [blame] | 58 | #### May 1 / November 1: Release freeze begins. |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 59 | |
| 60 | This milestone begins the second half of the release cycle, the release freeze. |
| 61 | The release freeze applies to the entire main repository |
| 62 | as well as to the code in subrepositories that is needed to build the |
| 63 | binaries included in the release, |
| 64 | particularly godoc and all its dependencies in the tools subrepository. |
| 65 | |
| 66 | Changes that were mailed before the freeze can be submitted |
| 67 | if they are reviewed promptly after the freeze begins. |
| 68 | During the freeze, only bug fixes and doc updates are accepted. |
| 69 | On occasion new work may be done during the freeze, but only in exceptional circumstances |
| 70 | and typically only if the work was proposed and approved before the cutoff. |
| 71 | Such changes must be low risk. |
| 72 | |
| 73 | This part of the release cycle is focused on improving the quality of the release, |
| 74 | by testing it and fixing bugs that are found. |
| 75 | However, every fix must be evaluated to balance the benefit of a possible fix |
| 76 | against the cost of now having not as well tested code (the fix) in the release. |
| 77 | Early in the release cycle, the balance tends toward accepting a fix. |
| 78 | Late in the release cycle, the balance tends toward rejecting a fix, |
| 79 | unless a case can be made that the fix is both low risk and high reward. |
| 80 | |
| 81 | Examples of low risk changes appropriate late in the cycle include changes |
| 82 | to documentation and fixes to new features being introduced in the current |
| 83 | release (since there is no chance of introducing a regression compared to |
| 84 | an earlier release). |
| 85 | |
| 86 | By the end of the first month of the freeze, nearly all known bugs |
| 87 | should have been fixed or explictly postponed (either to the next |
| 88 | release or indefinitely). There should be few known bugs left, |
| 89 | perhaps only the ones that have proven too elusive to chase down. |
| 90 | |
Russ Cox | 76b4923 | 2016-01-28 10:20:05 -0500 | [diff] [blame] | 91 | #### June 1 / December 1: Beta 1 issued. |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 92 | |
| 93 | A beta release is meant to encourage testing to discover new bugs. |
| 94 | Issuing a beta is an indication that the Go team has fixed nearly all the |
| 95 | known bugs that are planned to be fixed for the release, |
| 96 | and now it is time to look for as yet unknown bugs. |
| 97 | |
| 98 | The first beta includes a complete draft of the eventual release notes, |
| 99 | but marked clearly as a draft to avoid confusion when people link |
| 100 | to them on the internet. |
| 101 | |
| 102 | If a release is ahead of schedule, it is acceptable and even encouraged |
| 103 | to issue a beta a few weeks ahead of this target. |
| 104 | |
| 105 | As bugs are reported and fixed, additional betas may be released, |
| 106 | provided there are significant code changes to test anew. |
| 107 | Typically betas should not be issued more frequently than two weeks. |
| 108 | It is important not to issue too many betas, nor too many release candidates: |
| 109 | we are asking our users for their time to help us test the release |
| 110 | and must not waste their good will by making too many requests. |
| 111 | |
| 112 | A beta is not expected to be bug-free and should not be used in |
| 113 | production settings where failures or misbehavior cannot be tolerated. |
| 114 | Organizations can run integration or other tests against a beta |
| 115 | or even use it in a canary setting, but they should be discouraged from |
| 116 | deploying the beta for unrestricted production usage. |
| 117 | |
Russ Cox | 76b4923 | 2016-01-28 10:20:05 -0500 | [diff] [blame] | 118 | #### July 1 / January 1: Release candidate 1 issued. |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 119 | |
| 120 | A release candidate is meant to be as close as possible to the actual release bits. |
| 121 | Issuing a release candidate is an indication that the Go team has high confidence |
| 122 | that the tree is free of critical bugs. |
| 123 | |
| 124 | Once a release candidate is issued, only documentation changes |
| 125 | and changes to address critical bugs should be made. |
| 126 | In general the bar for bug fixes at this point is even slightly higher than the |
| 127 | bar for bug fixes in a minor release. |
| 128 | We may prefer to issue a release with a known but very rare crash |
| 129 | than to issue a release with a new but not production-tested fix. |
| 130 | |
| 131 | If a release is ahead of schedule, it is acceptable and even encouraged |
| 132 | to issue a release candidate a few weeks ahead of this target. |
| 133 | Extended release testing is a great way to deliver a robust release. |
| 134 | |
| 135 | If critical bugs are reported and fixed, additional release candidates may be issued, |
| 136 | but typically not more than one every two weeks. |
| 137 | |
| 138 | Again, a release candidate is meant to be bug-free, as much as possible. |
| 139 | Organizations are encouraged to deploy it in production settings |
| 140 | after appropriate organization-specific testing. |
| 141 | |
| 142 | One of the criteria for issuing a release candidate is that |
| 143 | Google be using that version of the code for new production builds |
| 144 | by default: if we at Google are not willing to run it for production use, |
| 145 | we shouldn't be asking others to. |
| 146 | We may issue the release candidate a few days in advance of Google |
| 147 | changing over, depending on how the calendar falls. |
| 148 | For example, the change inside Google makes more sense to do on Mondays, |
| 149 | so we may issue the release candidate the Wednesday before |
| 150 | or the Wednesday after Google converts to the new release by default. |
| 151 | |
| 152 | The calm period between a release candidate and the final release |
| 153 | is a good time for additional testing or for discussing the next release |
| 154 | (see the January 15 milestone above). |
| 155 | |
Russ Cox | 76b4923 | 2016-01-28 10:20:05 -0500 | [diff] [blame] | 156 | #### August 1 / February 1: Release issued. |
Russ Cox | 2ccff3f | 2016-01-28 10:16:44 -0500 | [diff] [blame] | 157 | |
| 158 | Finally, the release itself! |
| 159 | |
| 160 | A release should not contain significant changes since the last release candidate: |
| 161 | it is important that all code in the release has been well tested. |
| 162 | Issuing a release is an indication that release testing has confirmed |
| 163 | the release candidate's high confidence that the tree is free of critical bugs. |
| 164 | |
| 165 | One of the criteria for issuing a release is that the release candidate |
| 166 | has been available for four weeks and any problems that need to be |
| 167 | addressed have been. |
| 168 | |
| 169 | If a release process has run ahead of schedule, with an early beta |
| 170 | and early release candidate, |
| 171 | release candidate testing should absorb any extra time, |
| 172 | leaving the actual release on time, not early. |
| 173 | This improves the stability of the release, and it also gives developers |
| 174 | working on the Go release more time to think about and plan the |
| 175 | next release before code changes start pouring in again. |
| 176 | |
| 177 | If a release is behind schedule, it is acceptable (but certainly not ideal) |
| 178 | to issue a release sooner than four weeks after the release candidate, |
| 179 | but no sooner than two weeks after. |
| 180 | Abbreviated release testing is a great way to deliver a buggy release. |
| 181 | |
| 182 | Because Google runs the release candidate as the default |
| 183 | version of Go, four weeks of release testing means that |
| 184 | at the least Google has been using this version of Go for |
| 185 | four weeks before it becomes an official release. |
| 186 | While Google's successful use does not guarantee the |
| 187 | absence of problems, our experience has been that |
| 188 | it certainly helps improve the quality of the release. |
| 189 | We strongly encourage other organizations to test release candidates |
| 190 | as aggressively as they are able and to report problems that they find. |
| 191 | |
| 192 | Once a release is issued, work on the next release, including code reviews |
| 193 | and submission of new code, can begin, and the cycle repeats. |
| 194 | Note that if a release is delayed, so is work on the next release. |
| 195 | |
| 196 | ## Release Maintenance |
| 197 | |
| 198 | A minor release is issued to address one or more critical problem |
| 199 | for which there is no workaround (typically related to stability or security). |
| 200 | The only code changes included in the release are the fixes for the specific |
| 201 | critical problems. |
| 202 | Important documentation-only changes may also be included as well, |
| 203 | but nothing more. |
| 204 | |
| 205 | Minor releases to address non-security problems for Go 1.x stop once Go 1.x+1 is released. |
| 206 | |
| 207 | Minor releases to address security problems for Go 1.x stop once Go 1.x+2 is released. |
| 208 | For more about security updates, see the [security policy](https://golang.org/security). |
| 209 | |
| 210 | ## History |
| 211 | |
| 212 | The Go release cycle was discussed and adopted |
| 213 | after the fourteen month effort to release Go 1.1. |
| 214 | Go 1.2, Go 1.3, and Go 1.4 followed a six-month cycle |
| 215 | beginning and ending (alternately) on December 1 and June 1. |
| 216 | After experience with calendar problems in that cycle, |
| 217 | we extended Go 1.5's development phase by two months |
| 218 | to shift the cycle to begin and end on February 1 and August 1, |
| 219 | as described above. |
| 220 | |
| 221 | The [original proposal](https://golang.org/s/release-old) |
| 222 | did not contain enough detail about the milestones during the freeze, |
| 223 | and over the course of a few releases development work took over |
| 224 | much of the freeze. |
| 225 | Compared to the goal set above of issuing a beta one month |
| 226 | into the release freeze, the first betas for Go 1.3, Go 1.4, Go 1.5, |
| 227 | and Go 1.6 were three, four, five, and six weeks late, respectively. |
| 228 | (Go 1.6 beta 1 was only two weeks late, but it was |
| 229 | full of known bugs that we still intended to fix, |
| 230 | primarily to get something out for testing before the winter holidays. |
| 231 | Go 1.6's first real beta by the above definition was beta 2.) |
| 232 | |
| 233 | When the beta is late, everything that follows the beta—shaking out |
| 234 | the final bugs, thorough testing of the release candidates, |
| 235 | and the shipping of the release—gets rushed, leading to more bugs |
| 236 | in the final release and usually a delay in starting the next cycle. |
| 237 | |
| 238 | The beta was ready later and later in those four cycles primarily |
| 239 | because we both postponed too many bugs to the freeze |
| 240 | and then allowed too many non-essential bug fixes during the freeze. |
| 241 | |
| 242 | For Go 1.7 and later we will need to make sure that bugs are fixed |
| 243 | before the freeze. |
| 244 | That is, we need to follow the schedule above, not what we've done in the past. |