Github user ahgittin commented on a diff in the pull request:
https://github.com/apache/brooklyn-server/pull/867#discussion_r146482165
--- Diff:
core/src/main/java/org/apache/brooklyn/core/mgmt/ha/OsgiArchiveInstaller.java
---
@@ -379,8 +379,15 @@ private synchronized void close() {
if (canUpdate()) {
result.bundle =
osgiManager.framework.getBundleContext().getBundle(result.getMetadata().getOsgiUniqueUrl());
if (result.getBundle()==null) {
- log.warn("Brooklyn thought is was already managing
bundle "+result.getMetadata().getVersionedName()+" but it's not installed to
framework; reinstalling it");
- updating = false;
+ if (hasBundleOnInstall(osgiManager,
inferredMetadata, zipFile)) {
+ // e.g. happens if system-bundle is brought
under management, and then try to add it again:
+ // if added from "initial catalog" has it, and
then added from persisted state as well.
+ result.setIgnoringAlreadyInstalled();
+ return
ReferenceWithError.newInstanceWithoutError(result);
--- End diff --
This return path only takes effect if passing the `canUpdate()` check on
line 379 but presumably behaviour should be the same as for a non-snapshot
version?
Is the checksum check we do in the other block (line 397 in new money) is
equivalent or if we need a super-set check combining that line with
`hasBundleOnInstall()`.
Probably those checks and the two `ignoringAlreadyInstalled` returns should
be moved to before line 379.
I'd also rename the new `hasBundleOnInstall` method to
`isEquivalentBundleAlreadyOsgiInstalled` or something else a little more
descriptive.
---
