[SOLR-16813] PackageLoader not copying manifest.json file to filestore

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 9.2, 9.2.1
• Component/s: Package Manager
• Assignee: Unassigned
• Reporter: Will White
• Created: 22/May/23 16:13
• Updated: 21/Jul/23 20:42

Description

When a new Solr node enters the cluster and the package management system recreates the package from the source node, the filestore location is set up and the JAR files are copied across correctly, but the manifest.json file isn't (and the .manifest.json.json file isn't generated). This means that when trying to interact with the package (such as bin/solr package list-installed) the command errors out because of the missing expected file.

This can be reproduced by:

• Setup Zookeeper cluster
• Create two solr nodes ("solr1" and "solr2") in cluster
• Install a package on solr1
  • bin/solr package add-repo data-import-handler "https://raw.githubusercontent.com/searchscale/dataimporthandler/master/repo"
  • bin/solr package install data-import-handler
  • find /var/solr/data/filestore (manifest.json and .manifest.json.json should exist)
  • bin/solr package list-installed
• At this point, the package exists as expected on solr2
• Stop solr1 and solr2
• Restart the cluster, now with an additional solr3 node
• solr3 will have the JAR files and their .json files, but the manifest.json file won't be present
  • Running bin/solr package list-installed will throw an exception

---

The difference appears to be that the
org.apache.solr.packagemanager.RepositoryManager::installPackage method contains the following lines:

// org/apache/solr/packagemanager/RepositoryManager.java#182
if (release.manifest == null) {
  String manifestJson = PackageUtils.getFileFromJarsAsString(downloaded, "manifest.json");
  if (manifestJson == null) {
    throw new SolrException(ErrorCode.NOT_FOUND, "No manifest found for package: " + packageName + ", version: " + version);
  }
  release.manifest = getMapper().readValue(manifestJson, SolrPackage.Manifest.class);
}
String manifestJson = getMapper().writeValueAsString(release.manifest);
// We go on to write this file later

But in the equivalent space in org.apache.solr.pkg.SolrPackageLoader.Version(), the equivalent section to fetch the JAR files if missing doesn't reference the version.manifest anywhere:

// org/apache/solr/pkg/SolrPackageLoader.java#276
coreContainer.getPackageStoreAPI().validateFiles(version.files, true, s -> errs.add(s));
if (!errs.isEmpty()) {
  ...
}
for (String file : version.files) {
  paths.add(coreContainer.getPackageStoreAPI().getPackageStore().getRealpath(file));
}

Something like this commit solves the problem by manually copying the file across, but I'm not sure if this is the best approach to take.

I'm happy to put a full PR together, although I'd appreciate it if someone could confirm whether the manifest.json file should be transferred from the source node like this (and can be treated as mandatory), or if I've misunderstood how this should be working.