...
- Frank
- Clustering
- Issue with transaction timeouts
- Only seen in clustering, not seen on single node
- Replication - all caches replicated
- Distribution - specifies how many nodes should hold a given cache entry
- Deleted objects re-appear after a time
- Nodeservice and objectservice, path most start with a '/'
- createObject requires absolute path
- updatePropertiesGraph requires relative path
- Will document environment and errors so others can reproduce
- Will create tuning page under "Development"
- AuthZ issue
- Can resolve with wrapping as new exception
- Clustering
- Adam
- Still getting machine rebuilt
- Spoke with Martin Dow
- He will be getting in touch soon
- Suggestion to make pattern for accepting non-code contributions more clear
Thursday January 30
Friday January 31
- Frank
- A. Soroka
- Still rebuilding my workstation
- managed to finish detailing the remaining work to assemble a kernel API
- pursuing ongoing discussions with Martin Dow around the use of ontologies in Fedora 4
- hopefully, will manage to get back to "OSGi-fying" the indexer codebase
- Andrew
- Continuing work on https://www.pivotaltracker.com/story/show/64112028
Monday February 3
- Adam
- Working on OSGi
- Laptop is back in order
- Frank
- Clustering working
- Using UDP for discovery of nodes
- Using TCP for communication
- Bundling messages improves performance
- But still slow
- Adding hundreds of properties to a node is very slow
- Will be profiling single-node updates of properties
- Andrew
- Wrapping up two "federation" tickets
- Adding properties to federated objects
- Running fixity over federated objects
- Wrapping up two "federation" tickets
Tuesday February 4
- Frank
- 300 ms (single in cluster)
- 150 ms (single non-clustered)
- Help in disabling indexing in repository.json
- ISPN: switched to non-transactional mode
- exceptions are thrown
- ISPN: switched to async mode, did not see any performance improvement
- ISPN: pessimistic locking vs. optimistic locking did not show performance diff
- Will try bundling actions in a single transaction
- Adam
- OSGi going forward - Solr libraries
- Designing new "update" strategy
- One goal is to get sparql machinery out of the kernel
- Need to insure only valid updates are allowed
- Need to determine how to select an update strategy, from the REST API
- Do we only allow updates via SPARQL-Update?
Wednesday February 5
- Frank
- Still running into issues
- Works using fsync per write
- Creating one object with 4000 children took 370 seconds
- Slowed down with addition of more children, but children are all at the same level
- ISPN async config does not work at all
- Fails in all modes, timeouts or path-not-found exceptions
- When transaction spans too many ingests, run into not enough filehandles
- Adam
- Will be reviewing tickets
- Moving forward without roadblocks
- Has no cavities
- Andrew
- Finished Federation fixity and properties
- New ticket for Federation caching of SHA-1
- Getting deployment to futures6 running again
- Need Adam to review some tickets
Thursday February 6