When attempting to open a repository with +1M indexed nodes, with indexing rebuild-upon-startup="if_missing" (the default), an OOM is encountered. This is with ISPN eviction strategy="LIRS" max-entries="50", and Java heap 4GB.

This appears to be the result of the logic in org.modeshape.jcr.RepositoryQueryManager.reindexContent(boolean, boolean), which first reads the IDs of all 1M indexed node IDs into memory, and then crawls the entire repository in search of un-indexed nodes.

I suspect that rebuild-upon-startup="always" would also fail, although I have not tried it.

Also, note that in line 214 of RepositoryQueryManager.java, the entire contents of excludedKeysFromIndexing is logged (to debug). With millions of indexed nodes, this results in millions of IDs printed to the log, which is not desirable.