Grant report for period 11.2019-1.2020 - Karol Skocik
On meetup in Prague, a decision was made to implement a garbage
collector, which would trim the history of account movements and limit
the amount of data each user has to keep on disk.
Various approaches how to implement a garbage collector were tried,
and a version of 1-phase collector was documented in a GC research
paper. (https://github.com/aeternity/papers/blob/master/Garbage%20collector%20for%20account%20state%20data.pdf)
Work on querying of a registered name in naming system using a hash
was finalized and merged.
Our Sophia contract language was improved by introducing of
AENS.update function, which allows changing properties of a name and
updating of name pointers from contracts.
This fixed the feature parity gap we had in package covering naming
system functionality.
There was also checking of the viability of Lumen
(GitHub - GetFirefly/firefly: An alternative BEAM implementation, designed for WebAssembly) to bring Sophia compiler and parts of
the node to mobile and browser. While this approach shows a great
promise, Lumen needs to merge compiler branch to the main branch
before it can be useful for building deliverables running on other
platforms.
After assessing the features of 1-phase garbage collector and
discussion on Sofia meetup, we decided to implement a 2-phase
collector. While this version performs slower, it brings more
confidence regarding the stability of the node and preservation of all
user’s data.
Work on 2-phase collector was interrupted by middleware takeover
written in Rust. The middleware powers sites like aeternal.io and is
used by surrounding ecosystem (like airgap).
After middleware takeover, I picked up on remaining work on 2-phase
garbage collector - especially testing and stabilization of the
running garbage collector over longer period of time (4 days). The aim
was to gain confidence that running garbage collector in a loop
doesn’t case any node unstability and is ready for merging with the
rest of node codebase.
Below is a summary of this testing:
Running from 27.1.2020 - 31.1.2020
Number of garbage collections: 562
Number of generations to keep: 500
Average time of main scanning phase: 98.1 seconds
Average time of writing nodes from cache to disk: 4.5 seconds
Average time of swapping account nodes: 17.65 seconds
Number of hash_not_present_in_db errors: 3
Stability issues (node crashes): 0