Performance problem deserializing using BoltGraphClient

[simendsjo]

We've long known that deserializing was a slow part in Neo4jClient, but we have some operations where it's becoming a real challenge.

I notice CypherJsonDeserializer(String) here. Doesn't BoltGraphClient use the binary interface and should avoid Json alltogether? Another issue is that types are treversed many times -- are they not cached or cached at a too low level?

[cskardon]

Yes, you are right, there is a pr which covers this, but it causes some breaking changes and I need to fully understand it to pull it in

[simendsjo]

Yes, you are right, there is a pr which covers this, but it causes some breaking changes and I need to fully understand it to pull it in

Which PR? I can test it to see if it works for this case and report the performance difference.

[cskardon]

PR #275 -- I imagine it will need some work to get it to compile atm, or it might not!

I'd be interested to know how much of your code (if any) you had to change to use it..

[simendsjo]

I tried this now. I squashed it and then merged into master just to check. Getting it to compile was easy, but it deleted a file we've done bugfixes in, which accounts for two failing tests.

[cskardon]

@simendsjo How have you found the PR? Faster ? Did you have to do code changes?

[simendsjo]

I don't remember exactly what happened. I remember trying, and finding I had to make several changes. I think I might have rediscovered problems with a few fixes, which in retrospect might not have been 100% correct..?

Both 593e6ab and 7918f21 pops to mind, but memory might not serve me right here.

I started patching things, but quickly broke the HTTP client when the Bolt client started working, and quickly gave up as I suspected I had to use a lot of time understanding the existing code, the patch, and the Neo4j API.