Skip to content

Latest commit

 

History

History
47 lines (33 loc) · 3 KB

db_faq.md

File metadata and controls

47 lines (33 loc) · 3 KB
Raw

Database FAQ

How to read DB directly - not by Json-RPC/Graphql:

There are 2 options exist:

  1. call --private.api.addr there is grpc interface with low-level data access methods - can read any data in any order, etc... Interface is here: https://github.com/ledgerwatch/interfaces/blob/master/remote/kv.proto Go/C++/Rust libs already exist. Names of buckets and their format you can find in bucket.go You can do such calls by network.
  2. Read Erigon's db while Erigon is running - it's also ok - just need be careful - do not run too long read transactions (long read transactions do block free space in DB). Then your app will share with Erigon same OS-level PageCache where hot part of db stored. It may be great - if you read hot data (for example do incremental update of graph node) - then your reads will be super fast and almost never touch disk. But if you wanna read cold data - then your app will load cold data to PageCache and maybe evict some Erigon's hot data. Probably it will not be very dangerous - because your reads will happen once while Erigon will touch hot data often and OS's built-in LRU will understand which data is more Hot and keep it in RAM.

this 2 options ^ are exactly how RPCDaemon works with flags --private.api.addr and --datadir. One by using grpc interface, another by opening Erigon's db in read-only mode while Erigon running. But both this options are using RoKV (stands for read-only) kv_abstract.go interface. Option 1 using kv_remote.go to implement RoKV, option 2 using - kv_mdbx.go

Erigon uses MDBX storage engine. But most information on the Internet about LMDB is also valid for MDBX.

We have Go, Rust and C++ implementations of RoKV interface.

Rationale and Architecture of DB interface: ./../../ethdb/Readme.md

MDBX docs: erthink.github.io/libmdbx/ and https://github.com/erthink/libmdbx/blob/master/mdbx.h

How RAM used

Erigon will use all available RAM, but this RAM will not belong to Eroigon’s process. OS will own all this memory. And OS will maintain hot part of DB in RAM. If OS will need RAM for other programs or for second Erigon instance OS will manage all the work. This called PageCache. Erigon itself using under 2Gb. So, Erigon will benefit from more RAM and will use all RAM without re-configuration. Same PageCache can be used by other processes if they run on same machine by just opening same DB file. For example if RPCDaemon started with —datadir option - it will open db of Erigon and will use same PageCache (if data A already in RAM because it’s hot and RPCDaemon read it - then it read it from RAM not from Disk). Shared memory.

This also means - if you restart Erigon - PageCache will stay alive. Because it doesn’t belongs to Erigon.

After machine reboot - warmup takes ~10min - Erigon will getting faster during this time (but we don’t have special code to warmup) - Erigon will just do usual work.