|
| 1 | +"""Script to build and update the Jupyter WebApp image. |
| 2 | +
|
| 3 | +Requires python3 |
| 4 | +
|
| 5 | +hub CLI depends on an OAuth token with repo permissions: |
| 6 | +https://hub.github.com/hub.1.html |
| 7 | + * It will look for environment variable GITHUB_TOKEN |
| 8 | +""" |
| 9 | + |
| 10 | +import logging |
| 11 | +import os |
| 12 | +import re |
| 13 | +import tempfile |
| 14 | +import yaml |
| 15 | + |
| 16 | +import fire |
| 17 | +import git |
| 18 | +import httplib2 |
| 19 | + |
| 20 | +from kubeflow.testing import util # pylint: disable=no-name-in-module |
| 21 | + |
| 22 | +from containerregistry.client import docker_creds |
| 23 | +from containerregistry.client import docker_name |
| 24 | +from containerregistry.client.v2_2 import docker_http |
| 25 | +from containerregistry.client.v2_2 import docker_image as v2_2_image |
| 26 | +from containerregistry.transport import transport_pool |
| 27 | + |
| 28 | +class WebAppUpdater(object): # pylint: disable=useless-object-inheritance |
| 29 | + def __init__(self): |
| 30 | + self._last_commit = None |
| 31 | + |
| 32 | + def build_image(self, build_project, registry_project): |
| 33 | + """Build the image. |
| 34 | +
|
| 35 | + Args: |
| 36 | + build_project: GCP project used to build the image. |
| 37 | + registry_project: GCP project used to host the image. |
| 38 | + """ |
| 39 | + env = dict() |
| 40 | + env.update(os.environ) |
| 41 | + env["PROJECT"] = build_project |
| 42 | + env["REGISTRY_PROJECT"] = registry_project |
| 43 | + env["GIT_TAG"] = self._last_commit |
| 44 | + |
| 45 | + with tempfile.NamedTemporaryFile() as hf: |
| 46 | + name = hf.name |
| 47 | + env["OUTPUT"] = name |
| 48 | + web_dir = self._component_dir() |
| 49 | + util.run(["make", "build-gcb"], env=env, cwd=web_dir) |
| 50 | + |
| 51 | + # TODO(jlewi): We want to get the actual image produced by GCB. Right |
| 52 | + # now this is a bit brittle because we have multiple layers of substitution |
| 53 | + # e.g. in the Makefile and then the GCB YAML. |
| 54 | + # It might be better to parse the stdout of make-build-gcb to get the |
| 55 | + # GCB job name and then fetch the GCB info specifying the images. |
| 56 | + with open(name) as hf: |
| 57 | + data = yaml.load(hf) |
| 58 | + |
| 59 | + return data["image"] |
| 60 | + |
| 61 | + def _replace_parameters(self, lines, values): |
| 62 | + """Replace parameters in ksonnet text. |
| 63 | +
|
| 64 | + Args: |
| 65 | + lines: Lines of text |
| 66 | + values: A dictionary containing the names of parameters and the values |
| 67 | + to set them to. |
| 68 | +
|
| 69 | + Returns: |
| 70 | + lines: Modified lines |
| 71 | + old: Dictionary of old values for these parameters |
| 72 | + """ |
| 73 | + old = {} |
| 74 | + for i, line in enumerate(lines): |
| 75 | + # Split the line on white space |
| 76 | + pieces = re.findall(r'\S+', line) |
| 77 | + |
| 78 | + # Check if this line is a parameter |
| 79 | + # // @optionalParam image string gcr.io/myimage Some image |
| 80 | + if len(pieces) < 5: |
| 81 | + continue |
| 82 | + |
| 83 | + if pieces[0] != "//" or pieces[1] != "@optionalParam": |
| 84 | + continue |
| 85 | + |
| 86 | + param_name = pieces[2] |
| 87 | + if not param_name in values: |
| 88 | + continue |
| 89 | + |
| 90 | + old[param_name] = pieces[4] |
| 91 | + logging.info("Changing param %s from %s to %s", param_name, pieces[4], |
| 92 | + values[param_name]) |
| 93 | + pieces[4] = values[param_name] |
| 94 | + |
| 95 | + lines[i] = " ".join(pieces) |
| 96 | + |
| 97 | + return lines, old |
| 98 | + |
| 99 | + def update_prototype(self, image): |
| 100 | + """Update the prototype file. |
| 101 | +
|
| 102 | + Args: |
| 103 | + image: New image to set |
| 104 | +
|
| 105 | + Returns: |
| 106 | + prototype_file: The modified prototype file or None if the image is |
| 107 | + already up to date. |
| 108 | + """ |
| 109 | + values = {"image": image} |
| 110 | + |
| 111 | + |
| 112 | + prototype_file = os.path.join(self._root_dir(), |
| 113 | + "kubeflow/jupyter/prototypes", |
| 114 | + "jupyter-web-app.jsonnet") |
| 115 | + with open(prototype_file) as f: |
| 116 | + prototype = f.read().split("\n") |
| 117 | + |
| 118 | + new_lines, old_values = self._replace_parameters(prototype, values) |
| 119 | + |
| 120 | + if old_values["image"] == image: |
| 121 | + logging.info("Existing image was already the current image; %s", image) |
| 122 | + return None |
| 123 | + temp_file = prototype_file + ".tmp" |
| 124 | + with open(temp_file, "w") as w: |
| 125 | + w.write("\n".join(new_lines)) |
| 126 | + os.rename(temp_file, prototype_file) |
| 127 | + |
| 128 | + return prototype_file |
| 129 | + |
| 130 | + @property |
| 131 | + def last_commit(self): |
| 132 | + """Get the last commit of a change to the source for the jupyter-web-app.""" |
| 133 | + if not self._last_commit: |
| 134 | + # Get the hash of the last commit to modify the source for the Jupyter web |
| 135 | + # app image |
| 136 | + self._last_commit = util.run(["git", "log", "-n", "1", |
| 137 | + "--pretty=format:\"%h\"", |
| 138 | + "components/jupyter-web-app"], |
| 139 | + cwd=self._root_dir()).strip("\"") |
| 140 | + |
| 141 | + return self._last_commit |
| 142 | + |
| 143 | + def _find_remote_repo(self, repo, remote_url): # pylint: disable=no-self-use |
| 144 | + """Find the remote repo if it has already been added. |
| 145 | +
|
| 146 | + Args: |
| 147 | + repo: The git python repo object. |
| 148 | + remote_url: The URL of the remote repo e.g. |
| 149 | + [email protected]:jlewi/kubeflow.git |
| 150 | +
|
| 151 | + Returns: |
| 152 | + remote: git-python object representing the remote repo or none if it |
| 153 | + isn't present. |
| 154 | + """ |
| 155 | + for r in repo.remotes: |
| 156 | + for u in r.urls: |
| 157 | + if remote_url == u: |
| 158 | + return r |
| 159 | + |
| 160 | + return None |
| 161 | + |
| 162 | + def all(self, build_project, registry_project, remote_fork, |
| 163 | + add_github_host=False): # pylint: disable=too-many-statements,too-many-branches |
| 164 | + """Build the latest image and update the prototype. |
| 165 | +
|
| 166 | + Args: |
| 167 | + build_project: GCP project used to build the image. |
| 168 | + registry_project: GCP project used to host the image. |
| 169 | + remote_fork: Url of the remote fork. |
| 170 | + The remote fork used to create the PR; |
| 171 | + e.g. [email protected]:jlewi/kubeflow.git. currently only ssh is |
| 172 | + supported. |
| 173 | + add_github_host: If true will add the github ssh host to known ssh hosts. |
| 174 | + """ |
| 175 | + repo = git.Repo(self._root_dir()) |
| 176 | + util.maybe_activate_service_account() |
| 177 | + last_commit = self.last_commit |
| 178 | + |
| 179 | + # Ensure github.com is in the known hosts |
| 180 | + if add_github_host: |
| 181 | + output = util.run(["ssh-keyscan", "github.com"]) |
| 182 | + with open(os.path.join(os.getenv("HOME"), ".ssh", "known_hosts"), |
| 183 | + mode='a') as hf: |
| 184 | + hf.write(output) |
| 185 | + |
| 186 | + if not remote_fork. startswith( "[email protected]"): |
| 187 | + raise ValueError("Remote fork currently only supports ssh") |
| 188 | + |
| 189 | + remote_repo = self._find_remote_repo(repo, remote_fork) |
| 190 | + |
| 191 | + if not remote_repo: |
| 192 | + fork_name = remote_fork.split(":", 1)[-1].split("/", 1)[0] |
| 193 | + logging.info("Adding remote %s=%s", fork_name, remote_fork) |
| 194 | + remote_repo = repo.create_remote(fork_name, remote_fork) |
| 195 | + |
| 196 | + logging.info("Last change to components-jupyter-web-app was %s", last_commit) |
| 197 | + |
| 198 | + base = "gcr.io/{0}/jupyter-web-app".format(registry_project) |
| 199 | + |
| 200 | + # Check if there is already an image tagged with this commit. |
| 201 | + image = base + ":" + self.last_commit |
| 202 | + transport = transport_pool.Http(httplib2.Http) |
| 203 | + src = docker_name.from_string(image) |
| 204 | + creds = docker_creds.DefaultKeychain.Resolve(src) |
| 205 | + |
| 206 | + image_exists = False |
| 207 | + try: |
| 208 | + with v2_2_image.FromRegistry(src, creds, transport) as src_image: |
| 209 | + logging.info("Image %s exists; digest: %s", image, |
| 210 | + src_image.digest()) |
| 211 | + image_exists = True |
| 212 | + except docker_http.V2DiagnosticException as e: |
| 213 | + if e.status == 404: |
| 214 | + logging.info("%s doesn't exist", image) |
| 215 | + else: |
| 216 | + raise |
| 217 | + |
| 218 | + if not image_exists: |
| 219 | + logging.info("Building the image") |
| 220 | + image = self.build_image(build_project, registry_project) |
| 221 | + logging.info("Created image: %s", image) |
| 222 | + else: |
| 223 | + logging.info("Image %s already exists", image) |
| 224 | + |
| 225 | + # We should check what the current image is if and not update it |
| 226 | + # if its the existing image |
| 227 | + prototype_file = self.update_prototype(image) |
| 228 | + |
| 229 | + if not prototype_file: |
| 230 | + logging.info("Prototype not updated so not creating a PR.") |
| 231 | + return |
| 232 | + |
| 233 | + branch_name = "update_jupyter_{0}".format(last_commit) |
| 234 | + |
| 235 | + if repo.active_branch.name != branch_name: |
| 236 | + logging.info("Creating branch %s", branch_name) |
| 237 | + |
| 238 | + branch_names = [b.name for b in repo.branches] |
| 239 | + if branch_name in branch_names: |
| 240 | + logging.info("Branch %s exists", branch_name) |
| 241 | + util.run(["git", "checkout", branch_name], cwd=self._root_dir()) |
| 242 | + else: |
| 243 | + util.run(["git", "checkout", "-b", branch_name], cwd=self._root_dir()) |
| 244 | + |
| 245 | + logging.info("Add file %s to repo", prototype_file) |
| 246 | + repo.index.add([prototype_file]) |
| 247 | + repo.index.commit("Update the jupyter web app image to {0}".format(image)) |
| 248 | + |
| 249 | + util.run(["git", "push", "-f", remote_repo.name], cwd=self._root_dir()) |
| 250 | + |
| 251 | + self.create_pull_request(commit=last_commit) |
| 252 | + |
| 253 | + def create_pull_request(self, base="kubeflow:master", commit=None): |
| 254 | + """Create a pull request. |
| 255 | +
|
| 256 | + Args: |
| 257 | + base: The base to use. Defaults to "kubeflow:master". This should be |
| 258 | + in the form <GitHub OWNER>:<branch> |
| 259 | + """ |
| 260 | + # TODO(jlewi): Modeled on |
| 261 | + # https://github.com/kubeflow/examples/blob/master/code_search/docker/ks/update_index.sh |
| 262 | + # TODO(jlewi): We should use the GitHub API and check if there is an |
| 263 | + # existing open pull request. Or potentially just use the hub CLI. |
| 264 | + |
| 265 | + if not commit: |
| 266 | + commit = self.last_commit |
| 267 | + logging.info("No commit specified defaulting to %s", commit) |
| 268 | + |
| 269 | + pr_title = "[auto PR] Update the jupyter-web-app image to {0}".format(commit) |
| 270 | + |
| 271 | + # See hub conventions: |
| 272 | + # https://hub.github.com/hub.1.html |
| 273 | + # The GitHub repository is determined automatically based on the name |
| 274 | + # of remote repositories |
| 275 | + output = util.run(["hub", "pr", "list", "--format=%U;%t\n"], |
| 276 | + cwd=self._root_dir()) |
| 277 | + |
| 278 | + |
| 279 | + lines = output.splitlines() |
| 280 | + |
| 281 | + prs = {} |
| 282 | + for l in lines: |
| 283 | + n, t = l.split(";", 1) |
| 284 | + prs[t] = n |
| 285 | + |
| 286 | + if pr_title in prs: |
| 287 | + logging.info("PR %s already exists to update the Jupyter web app image " |
| 288 | + "to %s", prs[pr_title], commit) |
| 289 | + return |
| 290 | + |
| 291 | + with tempfile.NamedTemporaryFile(delete=False) as hf: |
| 292 | + hf.write(pr_title.encode()) |
| 293 | + message_file = hf.name |
| 294 | + |
| 295 | + # TODO(jlewi): -f creates the pull requests even if there are local changes |
| 296 | + # this was useful during development but we may want to drop it. |
| 297 | + util.run(["hub", "pull-request", "-f", "--base=" + base, "-F", |
| 298 | + message_file], |
| 299 | + cwd=self._root_dir()) |
| 300 | + |
| 301 | + def _root_dir(self): |
| 302 | + this_dir = os.path.dirname(__file__) |
| 303 | + return os.path.abspath(os.path.join(this_dir, "..", "..", "..", "..")) |
| 304 | + |
| 305 | + def _component_dir(self): |
| 306 | + return os.path.join(self._root_dir(), "components", "jupyter-web-app") |
| 307 | + |
| 308 | +if __name__ == '__main__': |
| 309 | + logging.basicConfig(level=logging.INFO, |
| 310 | + format=('%(levelname)s|%(asctime)s' |
| 311 | + '|%(pathname)s|%(lineno)d| %(message)s'), |
| 312 | + datefmt='%Y-%m-%dT%H:%M:%S', |
| 313 | + ) |
| 314 | + logging.getLogger().setLevel(logging.INFO) |
| 315 | + fire.Fire(WebAppUpdater) |
0 commit comments