github.com/badrootd/celestia-core@v0.0.0-20240305091328-aa4207a4b25d/RELEASES.md (about) 1 # Releases 2 3 CometBFT uses modified [semantic versioning](https://semver.org/) with each 4 release following a `vX.Y.Z` format. CometBFT is currently on major version 0 5 and uses the minor version to signal breaking changes. The `main` branch is 6 used for active development and thus it is not advisable to build against it. 7 8 The latest changes are always initially merged into `main`. Releases are 9 specified using tags and are built from long-lived "backport" branches that are 10 cut from `main` when the release process begins. Each release "line" (e.g. 11 0.34 or 0.33) has its own long-lived backport branch, and the backport branches 12 have names like `v0.34.x` or `v0.33.x` (literally, `x`; it is not a placeholder 13 in this case). CometBFT only maintains the last two releases at a time (the 14 oldest release is predominantly just security patches). 15 16 ## Backporting 17 18 As non-breaking changes land on `main`, they should also be backported to 19 these backport branches. 20 21 We use Mergify's [backport feature](https://mergify.io/features/backports) to 22 automatically backport to the needed branch. There should be a label for any 23 backport branch that you'll be targeting. To notify the bot to backport a pull 24 request, mark the pull request with the label corresponding to the correct 25 backport branch. For example, to backport to v0.38.x, add the label 26 `S:backport-to-v0.38.x`. Once the original pull request is merged, the bot will 27 try to cherry-pick the pull request to the backport branch. If the bot fails to 28 backport, it will open a pull request. The author of the original pull request 29 is responsible for solving the conflicts and merging the pull request. 30 31 ### Creating a backport branch 32 33 If this is the first release candidate for a minor version release, e.g. 34 v0.25.0, you get to have the honor of creating the backport branch! 35 36 Note that, after creating the backport branch, you'll also need to update the 37 tags on `main` so that `go mod` is able to order the branches correctly. You 38 should tag `main` with a "dev" tag that is "greater than" the backport 39 branches tags. Otherwise, `go mod` does not 'know' whether commits on `main` 40 come before or after the release. 41 42 In the following example, we'll assume that we're making a backport branch for 43 the 0.38.x line. 44 45 1. Start on `main` 46 47 2. Ensure that there is a [branch protection 48 rule](https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/defining-the-mergeability-of-pull-requests/managing-a-branch-protection-rule) for the 49 branch you are about to create (you will need admin access to the repository 50 in order to do this). 51 52 3. Create and push the backport branch: 53 54 ```sh 55 git checkout -b v0.38.x 56 git push origin v0.38.x 57 ``` 58 59 4. Create a PR to update the documentation directory for the backport branch. 60 61 We rewrite any URLs pointing to `main` to point to the backport branch, 62 so that generated documentation will link to the correct versions of files 63 elsewhere in the repository. The following files are to be excluded from this 64 search: 65 66 * [`README.md`](./README.md) 67 * [`CHANGELOG.md`](./CHANGELOG.md) 68 * [`UPGRADING.md`](./UPGRADING.md) 69 70 The following links are to always point to `main`, regardless of where they 71 occur in the codebase: 72 73 * `https://github.com/cometbft/cometbft/blob/main/LICENSE` 74 75 Be sure to search for all of the following links and replace `main` with your 76 corresponding branch label or version (e.g. `v0.38.x` or `v0.38`): 77 78 * `github.com/cometbft/cometbft/blob/main` -> 79 `github.com/cometbft/cometbft/blob/v0.38.x` 80 * `github.com/cometbft/cometbft/tree/main` -> 81 `github.com/cometbft/cometbft/tree/v0.38.x` 82 * `docs.cometbft.com/main` -> `docs.cometbft.com/v0.38` 83 84 Once you have updated all of the relevant documentation: 85 86 ```sh 87 # Create and push the PR. 88 git checkout -b update-docs-v038x 89 git commit -m "Update docs for v0.38.x backport branch." 90 git push -u origin update-docs-v038x 91 ``` 92 93 Be sure to merge this PR before making other changes on the newly-created 94 backport branch. 95 96 After doing these steps, go back to `main` and do the following: 97 98 1. Create a new workflow to run e2e nightlies for the new backport branch. (See 99 [e2e-nightly-main.yml][e2e] for an example.) 100 101 2. Add a new section to the Mergify config (`.github/mergify.yml`) to enable the 102 backport bot to work on this branch, and add a corresponding `backport-to-v0.38.x` 103 [label](https://github.com/cometbft/cometbft/labels) so the bot can be triggered. 104 105 3. Add a new section to the Dependabot config (`.github/dependabot.yml`) to 106 enable automatic update of Go dependencies on this branch. Copy and edit one 107 of the existing branch configurations to set the correct `target-branch`. 108 109 [e2e]: https://github.com/cometbft/cometbft/blob/main/.github/workflows/e2e-nightly-main.yml 110 111 ## Pre-releases 112 113 Before creating an official release, especially a minor release, we may want to 114 create an alpha or beta version, or release candidate (RC) for our friends and 115 partners to test out. We use git tags to create pre-releases, and we build them 116 off of backport branches, for example: 117 118 * `v0.38.0-alpha.1` - The first alpha release of `v0.38.0`. Subsequent alpha 119 releases will be numbered `v0.38.0-alpha.2`, `v0.38.0-alpha.3`, etc. 120 121 Alpha releases are to be considered the _most_ unstable of pre-releases, and 122 are most likely not yet properly QA'd. These are made available to allow early 123 adopters to start integrating and testing new functionality before we're done 124 with QA. 125 126 * `v0.38.0-beta.1` - The first beta release of `v0.38.0`. Subsequent beta 127 releases will be numbered `v0.38.0-beta.2`, `v0.38.0-beta.3`, etc. 128 129 Beta releases can be considered more stable than alpha releases in that we 130 will have QA'd them better than alpha releases, but there still may be 131 minor breaking API changes if users have strong demands for such changes. 132 133 * `v0.38.0-rc1` - The first release candidate (RC) of `v0.38.0`. Subsequent RCs 134 will be numbered `v0.38.0-rc2`, `v0.38.0-rc3`, etc. 135 136 RCs are considered more stable than beta releases in that we will have 137 completed our QA on them. APIs will most likely be stable at this point. The 138 difference between an RC and a release is that there may still be small 139 changes (bug fixes, features) that may make their way into the series before 140 cutting a final release. 141 142 (Note that branches and tags _cannot_ have the same names, so it's important 143 that these branches have distinct names from the tags/release names.) 144 145 If this is the first pre-release for a minor release, you'll have to make a new 146 backport branch (see above). Otherwise: 147 148 1. Start from the backport branch (e.g. `v0.38.x`). 149 2. Run the integration tests and the E2E nightlies 150 (which can be triggered from the GitHub UI; 151 e.g., <https://github.com/cometbft/cometbft/actions/workflows/e2e-nightly-37x.yml>). 152 3. Prepare the pre-release documentation: 153 * Build the changelog with [unclog] _without_ doing an unclog release, and 154 commit the built changelog. This ensures that all changelog entries appear 155 under an "Unreleased" heading in the pre-release's changelog. The changes 156 are only considered officially "released" once we cut a regular (final) 157 release. 158 * Ensure that `UPGRADING.md` is up-to-date and includes notes on any breaking 159 changes or other upgrading flows. 160 4. Prepare the versioning: 161 * Bump TMVersionDefault version in `version.go` 162 * Bump P2P and block protocol versions in `version.go`, if necessary. 163 Check the changelog for breaking changes in these components. 164 * Bump ABCI protocol version in `version.go`, if necessary 165 5. Open a PR with these changes against the backport branch. 166 6. Once these changes have landed on the backport branch, be sure to pull them back down locally. 167 7. Once you have the changes locally, create the new tag, specifying a name and a tag "message": 168 `git tag -a v0.38.0-rc1 -m "Release Candidate v0.38.0-rc1` 169 8. Push the tag back up to origin: 170 `git push origin v0.38.0-rc1` 171 Now the tag should be available on the repo's releases page. 172 9. Future pre-releases will continue to be built off of this branch. 173 174 ## Minor release 175 176 This minor release process assumes that this release was preceded by release 177 candidates. If there were no release candidates, begin by creating a backport 178 branch, as described above. 179 180 Before performing these steps, be sure the 181 [Minor Release Checklist](#minor-release-checklist) has been completed. 182 183 1. Start on the backport branch (e.g. `v0.38.x`) 184 2. Run integration tests (`make test_integrations`) and the e2e nightlies. 185 3. Prepare the release: 186 * Do a [release][unclog-release] with [unclog] for the desired version, 187 ensuring that you write up a good summary of the major highlights of the 188 release that users would be interested in. 189 * Build the changelog using unclog, and commit the built changelog. 190 * Ensure that `UPGRADING.md` is up-to-date and includes notes on any breaking changes 191 or other upgrading flows. 192 * Bump TMVersionDefault version in `version.go` 193 * Bump P2P and block protocol versions in `version.go`, if necessary 194 * Bump ABCI protocol version in `version.go`, if necessary 195 4. Open a PR with these changes against the backport branch. 196 5. Once these changes are on the backport branch, push a tag with prepared release details. 197 This will trigger the actual release `v0.38.0`. 198 * `git tag -a v0.38.0 -m 'Release v0.38.0'` 199 * `git push origin v0.38.0` 200 6. Make sure that `main` is updated with the latest `CHANGELOG.md`, `CHANGELOG_PENDING.md`, and `UPGRADING.md`. 201 202 ## Patch release 203 204 Patch releases are done differently from minor releases: They are built off of 205 long-lived backport branches, rather than from main. As non-breaking changes 206 land on `main`, they should also be backported into these backport branches. 207 208 Patch releases don't have release candidates by default, although any tricky 209 changes may merit a release candidate. 210 211 To create a patch release: 212 213 1. Checkout the long-lived backport branch: `git checkout v0.38.x` 214 2. Run integration tests (`make test_integrations`) and the nightlies. 215 3. Check out a new branch and prepare the release: 216 * Do a [release][unclog-release] with [unclog] for the desired version, 217 ensuring that you write up a good summary of the major highlights of the 218 release that users would be interested in. 219 * Build the changelog using unclog, and commit the built changelog. 220 * Bump the TMDefaultVersion in `version.go` 221 * Bump the ABCI version number, if necessary. (Note that ABCI follows semver, 222 and that ABCI versions are the only versions which can change during patch 223 releases, and only field additions are valid patch changes.) 224 4. Open a PR with these changes that will land them back on `v0.38.x` 225 5. Once this change has landed on the backport branch, make sure to pull it locally, then push a tag. 226 * `git tag -a v0.38.1 -m 'Release v0.38.1'` 227 * `git push origin v0.38.1` 228 6. Create a pull request back to main with the CHANGELOG & version changes from the latest release. 229 * Remove all `R:patch` labels from the pull requests that were included in the release. 230 * Do not merge the backport branch into main. 231 232 ## Minor Release Checklist 233 234 The following set of steps are performed on all releases that increment the 235 _minor_ version, e.g. v0.25 to v0.26. These steps ensure that CometBFT is well 236 tested, stable, and suitable for adoption by the various diverse projects that 237 rely on CometBFT. 238 239 ### Feature Freeze 240 241 Ahead of any minor version release of CometBFT, the software enters 'Feature 242 Freeze' for at least two weeks. A feature freeze means that _no_ new features 243 are added to the code being prepared for release. No code changes should be made 244 to the code being released that do not directly improve pressing issues of code 245 quality. The following must not be merged during a feature freeze: 246 247 * Refactors that are not related to specific bug fixes. 248 * Dependency upgrades. 249 * New test code that does not test a discovered regression. 250 * New features of any kind. 251 * Documentation or spec improvements that are not related to the newly developed 252 code. 253 254 This period directly follows the creation of the [backport 255 branch](#creating-a-backport-branch). The CometBFT team instead directs all 256 attention to ensuring that the existing code is stable and reliable. Broken 257 tests are fixed, flakey-tests are remedied, end-to-end test failures are 258 thoroughly diagnosed and all efforts of the team are aimed at improving the 259 quality of the code. During this period, the upgrade harness tests are run 260 repeatedly and a variety of in-house testnets are run to ensure CometBFT 261 functions at the scale it will be used by application developers and node 262 operators. 263 264 ### Nightly End-To-End Tests 265 266 The CometBFT team maintains [a set of end-to-end 267 tests](https://github.com/cometbft/cometbft/blob/main/test/e2e/README.md#L1) 268 that run each night on the latest commit of the project and on the code in the 269 tip of each supported backport branch. These tests start a network of 270 containerized CometBFT processes and run automated checks that the network 271 functions as expected in both stable and unstable conditions. During the feature 272 freeze, these tests are run nightly and must pass consistently for a release of 273 CometBFT to be considered stable. 274 275 ### Upgrade Harness 276 277 The CometBFT team is creating an upgrade test harness to exercise the workflow 278 of stopping an instance of CometBFT running one version of the software and 279 starting up the same application running the next version. To support upgrade 280 testing, we will add the ability to terminate the CometBFT process at specific 281 pre-defined points in its execution so that we can verify upgrades work in a 282 representative sample of stop conditions. 283 284 ### Large Scale Testnets 285 286 The CometBFT end-to-end tests run a small network (~10s of nodes) to exercise 287 basic consensus interactions. Real world deployments of CometBFT often have 288 over a hundred nodes just in the validator set, with many others acting as full 289 nodes and sentry nodes. To gain more assurance before a release, we will also 290 run larger-scale test networks to shake out emergent behaviors at scale. 291 292 Large-scale test networks are run on a set of virtual machines (VMs). Each VM is 293 equipped with 4 Gigabytes of RAM and 2 CPU cores. The network runs a very simple 294 key-value store application. The application adds artificial delays to different 295 ABCI calls to simulate a slow application. Each testnet is briefly run with no 296 load being generated to collect a baseline performance. Once baseline is 297 captured, a consistent load is applied across the network. This load takes the 298 form of 10% of the running nodes all receiving a consistent stream of two 299 hundred transactions per minute each. 300 301 During each test net, the following metrics are monitored and collected on each 302 node: 303 304 * Consensus rounds per height 305 * Maximum connected peers, Minimum connected peers, Rate of change of peer connections 306 * Memory resident set size 307 * CPU utilization 308 * Blocks produced per minute 309 * Seconds for each step of consensus (Propose, Prevote, Precommit, Commit) 310 * Latency to receive block proposals 311 312 For these tests we intentionally target low-powered host machines (with low core 313 counts and limited memory) to ensure we observe similar kinds of resource contention 314 and limitation that real-world deployments of CometBFT experience in production. 315 316 #### 200 Node Testnet 317 318 To test the stability and performance of CometBFT in a real world scenario, 319 a 200 node test network is run. The network comprises 5 seed nodes, 100 320 validators and 95 non-validating full nodes. All nodes begin by dialing 321 a subset of the seed nodes to discover peers. The network is run for several 322 days, with metrics being collected continuously. In cases of changes to performance 323 critical systems, testnets of larger sizes should be considered. 324 325 #### Rotating Node Testnet 326 327 Real-world deployments of CometBFT frequently see new nodes arrive and old 328 nodes exit the network. The rotating node testnet ensures that CometBFT is 329 able to handle this reliably. In this test, a network with 10 validators and 330 3 seed nodes is started. A rolling set of 25 full nodes are started and each 331 connects to the network by dialing one of the seed nodes. Once the node is able 332 to blocksync to the head of the chain and begins producing blocks using 333 consensus it is stopped. Once stopped, a new node is started and 334 takes its place. This network is run for several days. 335 336 #### Network Partition Testnet 337 338 CometBFT is expected to recover from network partitions. A partition where no 339 subset of the nodes is left with the super-majority of the stake is expected to 340 stop making blocks. Upon alleviation of the partition, the network is expected 341 to once again become fully connected and capable of producing blocks. The 342 network partition testnet ensures that CometBFT is able to handle this 343 reliably at scale. In this test, a network with 100 validators and 95 full 344 nodes is started. All validators have equal stake. Once the network is 345 producing blocks, a set of firewall rules is deployed to create a partitioned 346 network with 50% of the stake on one side and 50% on the other. Once the 347 network stops producing blocks, the firewall rules are removed and the nodes 348 are monitored to ensure they reconnect and that the network again begins 349 producing blocks. 350 351 #### Absent Stake Testnet 352 353 CometBFT networks often run with _some_ portion of the voting power offline. 354 The absent stake testnet ensures that large networks are able to handle this 355 reliably. A set of 150 validator nodes and three seed nodes is started. The set 356 of 150 validators is configured to only possess a cumulative stake of 67% of 357 the total stake. The remaining 33% of the stake is configured to belong to 358 a validator that is never actually run in the test network. The network is run 359 for multiple days, ensuring that it is able to produce blocks without issue. 360 361 [unclog]: https://github.com/informalsystems/unclog 362 [unclog-release]: https://github.com/informalsystems/unclog#releasing-a-new-versions-change-set