ci-github: Re-request check suite on internal CI cancellation

4 files changed, 225 insertions, 2 deletions

diff --git a/mod/mod-ci-github-gq.cxx b/mod/mod-ci-github-gq.cxx
index 24f8b47..9d20c9e 100644
--- a/mod/mod-ci-github-gq.cxx
+++ b/mod/mod-ci-github-gq.cxx
@@ -1038,6 +1038,131 @@ namespace brep
     return r;
   }
 
+
+  // Serialize a GraphQL query that rerequests a check suite.
+  //
+  // Throw invalid_argument if any of the node ids are not a valid GraphQL
+  // string.
+  //
+  static string
+  gq_mutation_rerequest_check_suite (const string& rid, const string& nid)
+  {
+    ostringstream os;
+
+    os << "mutation {"                                                   << '\n'
+       << "  rerequestCheckSuite(input: {repositoryId: " << gq_str (rid) << '\n'
+       << "                              checkSuiteId: " << gq_str (nid) << '\n'
+       << "                      }) {"                                   << '\n'
+       << "    checkSuite { id }"                                        << '\n'
+       << "  }"                                                          << '\n'
+       << "}";
+
+    return os.str ();
+  }
+
+  bool
+  gq_rerequest_check_suite (const basic_mark& error,
+                            const string& iat,
+                            const string& rid,
+                            const string& nid)
+  {
+    // Let invalid_argument from gq_mutation_rerequest_check_suite()
+    // propagate.
+    //
+    string rq (
+      gq_serialize_request (gq_mutation_rerequest_check_suite (rid, nid)));
+
+    try
+    {
+      // Response parser.
+      //
+      struct resp
+      {
+        // True if the check suite was found (i.e., the node id was valid).
+        //
+        bool found = false;
+
+        // Example response (note: outer `data` object already stripped at
+        // this point):
+        //
+        // {"rerequestCheckSuite":{"checkSuite":{"id":"CS_kwDOLc8CoM8AAAAIDgO-Qw"}}}
+        //
+        resp (json::parser& p)
+        {
+          using event = json::event;
+
+          gq_parse_response (p, [this] (json::parser& p)
+          {
+            p.next_expect (event::begin_object); // Outer {
+
+            // This object will be null if the repository or check suite node
+            // ids were invalid.
+            //
+            if (p.next_expect_member_object_null ("rerequestCheckSuite"))
+            {
+              found = true;
+
+              p.next_expect_member_object ("checkSuite");
+              p.next_expect_member_string ("id");
+              p.next_expect (event::end_object); // checkSuite
+              p.next_expect (event::end_object); // rerequestCheckSuite
+            }
+
+            p.next_expect (event::end_object); // Outer }
+          });
+        }
+
+        resp () = default;
+      } rs;
+
+      uint16_t sc (github_post (rs,
+                                "graphql", // API Endpoint.
+                                strings {"Authorization: Bearer " + iat},
+                                move (rq)));
+
+      if (sc == 200)
+      {
+        if (!rs.found)
+        {
+          error << "check suite '" << nid
+                << "' not found in repository '" << rid << '\'';
+          return false;
+        }
+
+        return true;
+      }
+      else
+        error << "failed to re-request check suite: error HTTP response status "
+              << sc;
+    }
+    catch (const json::invalid_json_input& e) // struct resp (via github_post())
+    {
+      // Note: e.name is the GitHub API endpoint.
+      //
+      error << "malformed JSON in response from " << e.name << ", line: "
+            << e.line << ", column: " << e.column << ", byte offset: "
+            << e.position << ", error: " << e;
+    }
+    catch (const invalid_argument& e) // github_post()
+    {
+      error << "malformed header(s) in response: " << e;
+    }
+    catch (const system_error& e) // github_post()
+    {
+      error << "unable to re-request check suite (errno=" << e.code () << "): "
+            << e.what ();
+    }
+    catch (const runtime_error& e) // struct resp
+    {
+      // GitHub response contained error(s) (could be ours or theirs at this
+      // point).
+      //
+      error << "unable to re-request check suite: " << e;
+    }
+
+    return false;
+  }
+
   // Serialize a GraphQL query that fetches a pull request from GitHub.
   //
   // Throw invalid_argument if the node id is not a valid GraphQL string.
diff --git a/mod/mod-ci-github-gq.hxx b/mod/mod-ci-github-gq.hxx
index 77b78e4..4e13606 100644
--- a/mod/mod-ci-github-gq.hxx
+++ b/mod/mod-ci-github-gq.hxx
@@ -124,6 +124,21 @@ namespace brep
                        const string& node_id,
                        gq_built_result);
 
+  // Re-request a check suite. This will result in the delivery of a
+  // check_suite webhook with the "rerequested" action, just as if the user
+  // had clicked "re-run all checks" in the GitHub UI.
+  //
+  // Return false and issue diagnostics if the request failed.
+  //
+  // Throw invalid_argument if the passed data is invalid, missing, or
+  // inconsistent.
+  //
+  bool
+  gq_rerequest_check_suite (const basic_mark& error,
+                            const string& installation_access_token,
+                            const string& repository_id,
+                            const string& node_id);
+
   // Fetch pre-check information for a pull request from GitHub. This
   // information is used to decide whether or not to CI the PR and is
   // comprised of the PR's head commit SHA, whether its head branch is behind
diff --git a/mod/mod-ci-github.cxx b/mod/mod-ci-github.cxx
index 8f0af2e..c59b1af 100644
--- a/mod/mod-ci-github.cxx
+++ b/mod/mod-ci-github.cxx
@@ -1034,8 +1034,9 @@ namespace brep
     // Replace the existing CI tenant if it exists.
     //
     // Note that GitHub UI does not allow re-running the entire check suite
-    // until all the check runs are completed.
-    //
+    // until all the check runs are completed. However if we got here as a
+    // result of re-requesting the check suite from build_canceled() the check
+    // runs could be in any state.
 
     // Create an unloaded CI tenant.
     //
@@ -3213,6 +3214,81 @@ namespace brep
     error << "unhandled exception: " << e.what ();
   }
 
+  void ci_github::
+  build_canceled (const string& /* tenant_id */,
+                  const tenant_service& ts,
+                  const diag_epilogue& log_writer) const noexcept
+  try
+  {
+    // NOTE: this function is noexcept and should not throw.
+
+    NOTIFICATION_DIAG (log_writer);
+
+    // We end up here when the service data could not be saved so re-request
+    // the check suite in order to restart all of the builds.
+
+    // Load the unsaved service data.
+    //
+    service_data sd;
+    try
+    {
+      sd = service_data (*ts.data);
+    }
+    catch (const invalid_argument& e)
+    {
+      error << "failed to parse service data: " << e;
+      return;
+    }
+
+    // Get a new installation access token if the current one has expired.
+    //
+    const gh_installation_access_token* iat (nullptr);
+    optional<gh_installation_access_token> new_iat;
+
+    if (system_clock::now () > sd.installation_access.expires_at)
+    {
+      if (optional<string> jwt = generate_jwt (sd.app_id, trace, error))
+      {
+        new_iat = obtain_installation_access_token (sd.installation_id,
+                                                    move (*jwt),
+                                                    error);
+        if (new_iat)
+          iat = &*new_iat;
+      }
+    }
+    else
+      iat = &sd.installation_access;
+
+    if (iat != nullptr)
+    {
+      // Re-request the check suite.
+
+      // Note that the conclusion check run is created before the tenant is
+      // loaded so the unsaved service data should always contain the check
+      // suite node id.
+      //
+      assert (sd.check_suite_node_id.has_value ());
+
+      // Let unlikely invalid_argument propagate.
+      //
+      if (gq_rerequest_check_suite (error,
+                                    iat->token,
+                                    sd.repository_node_id,
+                                    *sd.check_suite_node_id))
+      {
+        l3 ([&]{trace << "re-requested check suite " << *sd.check_suite_node_id;});
+      }
+      else
+        error << "failed to re-request check suite " << *sd.check_suite_node_id;
+    }
+  }
+  catch (const std::exception& e)
+  {
+    NOTIFICATION_DIAG (log_writer);
+
+    error << "unhandled exception: " << e.what ();
+  }
+
   string ci_github::
   details_url (const build& b) const
   {
diff --git a/mod/mod-ci-github.hxx b/mod/mod-ci-github.hxx
index 0c90bb1..9a271aa 100644
--- a/mod/mod-ci-github.hxx
+++ b/mod/mod-ci-github.hxx
@@ -84,6 +84,13 @@ namespace brep
                      const tenant_service& ts,
                      const diag_epilogue& log_writer) const noexcept override;
 
+    // @@ TODO
+    //virtual void
+    void
+    build_canceled (const string& tenant_id,
+                    const tenant_service&,
+                    const diag_epilogue& log_writer) const noexcept; //override;
+
   private:
     virtual void
     init (cli::scanner&);