Add BatchPreload to decode slabs in parallel and cache

[fxamacker]

Updates #394

This PR adds BatchPreload which decodes slabs in parallel and stores decoded slabs in cache for later retrieval.

Migration benchmark (not available yet) will use https://github.com/onflow/flow-go/tree/feature/atree-inlining-cadence-v0.42

Casual microbenchmark (on my busy desktop):

                           │  before.txt  │              after.txt               │
                           │    sec/op    │    sec/op     vs base                │
StorageRetrieve/10-12         36.23µ ± 3%   35.33µ ±  4%        ~ (p=0.075 n=10)
StorageRetrieve/100-12        469.6µ ± 8%   124.3µ ±  0%  -73.52% (p=0.000 n=10)
StorageRetrieve/1000-12       6.678m ± 7%   2.303m ± 20%  -65.51% (p=0.000 n=10)
StorageRetrieve/10000-12      29.81m ± 2%   12.26m ±  5%  -58.86% (p=0.000 n=10)
StorageRetrieve/100000-12    303.33m ± 1%   88.40m ±  1%  -70.86% (p=0.000 n=10)
StorageRetrieve/1000000-12     3.442 ± 1%    1.137 ±  3%  -66.96% (p=0.000 n=10)
geomean                       12.34m        4.816m        -60.98%

                           │  before.txt  │              after.txt              │
                           │     B/op     │     B/op      vs base               │
StorageRetrieve/10-12        21.59Ki ± 0%   21.59Ki ± 0%       ~ (p=1.000 n=10)
StorageRetrieve/100-12       219.8Ki ± 0%   224.7Ki ± 0%  +2.24% (p=0.000 n=10)
StorageRetrieve/1000-12      2.266Mi ± 0%   2.272Mi ± 0%  +0.27% (p=0.000 n=10)
StorageRetrieve/10000-12     21.94Mi ± 0%   22.14Mi ± 0%  +0.91% (p=0.000 n=10)
StorageRetrieve/100000-12    215.3Mi ± 0%   218.5Mi ± 0%  +1.50% (p=0.000 n=10)
StorageRetrieve/1000000-12   2.211Gi ± 0%   2.212Gi ± 0%  +0.05% (p=0.000 n=10)
geomean                      6.919Mi        6.976Mi       +0.82%

                           │ before.txt  │              after.txt               │
                           │  allocs/op  │  allocs/op   vs base                 │
StorageRetrieve/10-12         76.00 ± 0%    76.00 ± 0%       ~ (p=1.000 n=10) ¹
StorageRetrieve/100-12        745.0 ± 0%    759.0 ± 0%  +1.88% (p=0.000 n=10)
StorageRetrieve/1000-12      7.161k ± 0%   7.153k ± 0%  -0.11% (p=0.000 n=10)
StorageRetrieve/10000-12     70.77k ± 0%   70.58k ± 0%  -0.27% (p=0.000 n=10)
StorageRetrieve/100000-12    711.9k ± 0%   709.7k ± 0%  -0.31% (p=0.000 n=10)
StorageRetrieve/1000000-12   7.115M ± 0%   7.077M ± 0%  -0.54% (p=0.000 n=10)
geomean                      22.93k        22.95k       +0.11%

---

• Targeted PR against main branch
• Linked to Github issue with discussion and accepted design OR link to spec that describes this work
• Code follows the standards mentioned here
• Updated relevant documentation
• Re-reviewed Files changed in the Github PR explorer
• Added appropriate labels

[fxamacker]

Looks like validation and storage health checks passed tonight with this PR added to atree migration program at:

• (locally modified) https://github.com/onflow/flow-go/tree/feature/atree-inlining-cadence-v0.42

Need to take another look at logs to confirm.

[fxamacker]

@ramtinms @turbolent PTAL 🙏

[ramtinms]

Looks good to me, only consideration is maybe just add some comments or warning about when to use this method.

[turbolent]

Great idea and nice work! 👏

[turbolent]

@fxamacker Does this also have to get ported to the 1.0 branch?

[fxamacker]

@fxamacker Does this also have to get ported to the 1.0 branch?

@turbolent I will port it for completeness, but it is only needed if we use the optimization for Cadence 1.0 migration without atree inlining.