[run_isolated] Tolerate cache uninstall errors.

client/run_isolated.py

Index: client/run_isolated.py
diff --git a/client/run_isolated.py b/client/run_isolated.py
index 8bbb6d233380d96d1fe035ca500b7e65c96fa877..3e035688f48eec2d5adcfe7b89bd9687f89e8355 100755
--- a/client/run_isolated.py
+++ b/client/run_isolated.py
@@ -1053,9 +1053,19 @@ def main(args):
     try:
       yield
     finally:
+      # Uninstall each named cache, returning it to the cache pool. If an
+      # uninstall fails for a given cache, it will remain in the task's
+      # temporary space, get cleaned up by the Swarming bot, and be lost.
+      #
+      # If the Swarming bot cannot clean up the cache, it will handle it like
+      # any other bot file that could not be removed.
       with named_cache_manager.open():
         for path, name in caches:
-          named_cache_manager.uninstall(path, name)
+          try:
+            named_cache_manager.uninstall(path, name)

[Vadim Sh., 2017/07/07 18:01:46]

does this thing leave 'named_cache_manager' in consistent state when an
exception happens?

[dnj, 2017/07/07 18:14:12]

Good question. It seems to:
https://cs.chromium.org/chromium/infra/luci/client/named_cache.py?q=named_cac...

If the "rename" succeeds but the LRU isn't updated, a cache entry won't be
present in the LRU. Next time the cache is referenced, it will be re-used and
added to the LRU next time (assuming it succeeds). "trim" also is resilient to
directories that aren't in LRU, as it does a "listdir".

All of the other operations are filesystem state, so nothing new is introduced
here.

+          except named_cache.Error:
+            logging.exception('Error while removing named cache %r at %r. '
+                              'The cache will be lost.', path, name)
 
   try:
     if options.isolate_server: