Attempt to more gracefully handle the problem of missing derivations

In the agent and allocator.

2 files changed, 44 insertions, 20 deletions

diff --git a/guix-build-coordinator/agent.scm b/guix-build-coordinator/agent.scm
index bb8d84b..96611fc 100644
--- a/guix-build-coordinator/agent.scm
+++ b/guix-build-coordinator/agent.scm
@@ -243,26 +243,40 @@
   (let ((derivation
          (if (valid-path? store derivation-name)
              (read-derivation-from-file derivation-name)
-             (begin (retry-on-error
-                     (lambda ()
-                       (substitute-derivation derivation-name
-                                              #:substitute-urls
-                                              derivation-substitute-urls))
-                     #:times 12
-                     #:delay 20)
-                    (read-derivation-from-file derivation-name)))))
-    (match (delete-outputs derivation)
-      (#t
-       (let ((missing-inputs
-              (find-missing-inputs derivation (derivation-inputs derivation))))
-         (if (null? missing-inputs)
-             '((result . success))
-             `((result         . failure)
-               (failure_reason . missing_inputs)
-               (missing_inputs . ,(list->vector missing-inputs))))))
-      (failure
-       '((result         . failure)
-         (failure_reason . could_not_delete_outputs))))))
+             (and
+              (with-exception-handler
+                  (lambda (exn)
+                    (simple-format
+                     (current-error-port)
+                     "error: exception when reading/fetching derivation: ~A\n"
+                     exn)
+                    #f)
+                (lambda ()
+                  (retry-on-error
+                   (lambda ()
+                     (substitute-derivation derivation-name
+                                            #:substitute-urls
+                                            derivation-substitute-urls))
+                   #:times 12
+                   #:delay 20)
+                  #t)
+                #:unwind? #t)
+              (read-derivation-from-file derivation-name)))))
+    (if derivation
+        (match (delete-outputs derivation)
+          (#t
+           (let ((missing-inputs
+                  (find-missing-inputs derivation (derivation-inputs derivation))))
+             (if (null? missing-inputs)
+                 '((result . success))
+                 `((result         . failure)
+                   (failure_reason . missing_inputs)
+                   (missing_inputs . ,(list->vector missing-inputs))))))
+          (failure
+           '((result         . failure)
+             (failure_reason . could_not_delete_outputs))))
+        '((result         . failure)
+          (failure_reason . error_fetching_derivation)))))
 
 (define (perform-build store derivation-name)
   (set-build-options store
diff --git a/guix-build-coordinator/build-allocator.scm b/guix-build-coordinator/build-allocator.scm
index f4de8b7..1b7bd0b 100644
--- a/guix-build-coordinator/build-allocator.scm
+++ b/guix-build-coordinator/build-allocator.scm
@@ -89,6 +89,11 @@
             ;; again for now.
             (string=? (assq-ref setup-failure 'agent-id)
                       agent-id))
+           ((string=? failure-reason "error_fetching_derivation")
+            ;; This problem might go away, but just don't try the same agent
+            ;; again for now.
+            (string=? (assq-ref setup-failure 'agent-id)
+                      agent-id))
            (else
             (error "Unknown setup failure " failure-reason)))))
 
@@ -380,6 +385,11 @@
               ;; again for now.
               (string=? (assq-ref setup-failure 'agent-id)
                         agent-id))
+             ((string=? failure-reason "error_fetching_derivation")
+              ;; This problem might go away, but just don't try the same agent
+              ;; again for now.
+              (string=? (assq-ref setup-failure 'agent-id)
+                        agent-id))
              (else
               (error "Unknown setup failure " failure-reason)))))