Add new flag for maximum file descriptors in rocksdb. #2386

AurelienFT · 2024-10-23T10:09:06Z

Linked Issues/PRs

Resolves: #2093

Description

Add a flag to define the maximum number of file descriptors that RocksDB can use. By default it's half of the OS limit.

Checklist

Breaking changes are clearly marked as such in the PR description and changelog
New behavior is reflected in tests
The specification matches the implemented behavior (link update PR if changes are needed)

Before requesting review

I have reviewed the code myself
I have created follow-up issues caused by this PR and linked them here

MitchTurner

Looking good. Left some thoughts.

bin/fuel-core/src/cli/run.rs

crates/fuel-core/src/state/rocks_db.rs

…o limit_rocks_db_fd

MitchTurner · 2024-10-23T15:23:52Z

bin/fuel-core/src/cli/rollback.rs

@@ -30,7 +30,7 @@ pub struct Command {
    #[clap(
        long = "rocksdb-max-fds",
        env,
-        default_value = getrlimit(Resource::NOFILE).map(|(_, hard)| i32::try_from(hard.saturating_div(2)).unwrap_or(i32::MAX)).unwrap().to_string()
+        default_value = getrlimit(Resource::NOFILE).map(|(_, hard)| i32::try_from(hard.saturating_div(2)).unwrap_or(i32::MAX)).expect("Failed to get OS max file descriptors.").to_string()


Oh. Sorry. I mean specify why it shouldn't panic:

We recommend that expect messages are used to describe the reason you expect the Result should be Ok.

As suggested by the docs:
https://doc.rust-lang.org/std/result/enum.Result.html#method.expect

Sorry I didn't correctly understood you first message and didn't know about this convention. Happy to learn :D Ty

Oh didn't know this. I've definitely done a bunch of .expect("failed to blah blah") 😅 TIL

netrome

Looks correct to me. I think this would be even better if we used an Option<u32> instead of an i32 where -1 has a special meaning.

netrome · 2024-10-24T08:12:12Z

bin/fuel-core/src/cli/rollback.rs

+    /// If defined as -1 no limit will be applied and will use the OS limits.
+    /// If not defined the system default divided by two is used.


What's the reason for using -1 to mean no limit as opposed to take an Option<u32>?

It is the default value used by RocksDB for no limit that's why they are a i32.

And we want to have different behaviour between :

No defined value which will be using OS divided by two

and max value of the system which is -1

Alright, fair enough.

I agree that this limit should generally be Option<u32>, but this can only represent two states, not the three we need. Let's keep it like that for the sake of simplicity. But we should maybe sanitize this value and not allow -2, -3, etc.?

RocksDB doesn't seems to consider these values as error : #2386 (comment)

But we can do. I don't think it's very useful because it doesn't creates any problem and the only reason to tryy a -2 is to tryy to make a problem.

netrome · 2024-10-24T08:12:49Z

bin/fuel-core/src/cli/rollback.rs

+    #[clap(
+        long = "rocksdb-max-fds",
+        env,
+        default_value = getrlimit(Resource::NOFILE).map(|(_, hard)| i32::try_from(hard.saturating_div(2)).unwrap_or(i32::MAX)).expect("Our supported platforms should return max FD.").to_string()


This is quite long to be in an attribute. How about breaking this out to a helper function?

netrome · 2024-10-24T08:13:23Z

bin/fuel-core/src/cli/run.rs

+    #[clap(
+        long = "rocksdb-max-fds",
+        env,
+        default_value = getrlimit(Resource::NOFILE).map(|(_, hard)| i32::try_from(hard.saturating_div(2)).unwrap_or(i32::MAX)).expect("Our supported platforms should return max FD.").to_string()


netrome · 2024-10-24T08:13:38Z

bin/fuel-core/src/cli/snapshot.rs

+    #[clap(
+        long = "rocksdb-max-fds",
+        env,
+        default_value = getrlimit(Resource::NOFILE).map(|(_, hard)| i32::try_from(hard.saturating_div(2)).unwrap_or(i32::MAX)).expect("Our supported platforms should return max FD.").to_string()


MitchTurner · 2024-10-24T09:01:26Z

Ready to approve after those last changes are made.

MitchTurner · 2024-10-24T09:16:33Z

benches/benches/vm_set/blockchain.rs

@@ -75,6 +75,7 @@ impl BenchDb {
            tmp_dir.path(),
            None,
            Default::default(),
+            -1,


There are a few places that we're still using -1. Is that okay?

I have let in benches because benches didn't used the test value before and I don't want them to be slowed down by a new limit (maybe it don't change anything but there is no risk for benches I think)

MitchTurner · 2024-10-24T09:16:57Z

bin/fuel-core/src/cli/rollback.rs

@@ -19,11 +23,28 @@ pub struct Command {
    )]
    pub database_path: PathBuf,

+    /// Defines a specific number of file descriptors that RocksDB can use.
+    ///
+    /// If defined as -1 no limit will be applied and will use the OS limits.


What is the behavior if they use -2 or any other negative value?

Didn't found anything about it in rocksdb source code in Rust or C, so I tried myself in a test and it seems to work without issues. Maybe this is the code : https://github.com/facebook/rocksdb/blob/8b38d4b4006ca9fd49432ccc16d9911919870dd2/db/db_impl/db_impl_open.cc#L57 I'm not sure that I fully understand this code but fore me if the value is -2 it will be clamped to 20.

Add new flag for maximum file descriptors in rocksdb.

87a2240

AurelienFT requested review from xgreenx, Dentosal and MitchTurner as code owners October 23, 2024 10:09

Merge branch 'master' into limit_rocks_db_fd

ef6a1d0

MitchTurner reviewed Oct 23, 2024

View reviewed changes

bin/fuel-core/src/cli/run.rs Outdated Show resolved Hide resolved

crates/fuel-core/src/state/rocks_db.rs Outdated Show resolved Hide resolved

AurelienFT added 2 commits October 23, 2024 16:01

Replace unwrap with expect and change vlaue in tests.

929f0bd

Merge branch 'limit_rocks_db_fd' of github.com:FuelLabs/fuel-core int…

09d4ddd

…o limit_rocks_db_fd

AurelienFT requested a review from MitchTurner October 23, 2024 14:02

MitchTurner reviewed Oct 23, 2024

View reviewed changes

Update expect message.

82eaa5f

AurelienFT requested review from MitchTurner and a team October 23, 2024 19:20

AurelienFT self-assigned this Oct 24, 2024

netrome previously approved these changes Oct 24, 2024

View reviewed changes

Update default function to helper.

c1bec23

AurelienFT dismissed netrome’s stale review via c1bec23 October 24, 2024 09:11

MitchTurner reviewed Oct 24, 2024

View reviewed changes

Merge branch 'master' into limit_rocks_db_fd

9d41bd5

MitchTurner approved these changes Oct 25, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add new flag for maximum file descriptors in rocksdb. #2386

Add new flag for maximum file descriptors in rocksdb. #2386

AurelienFT commented Oct 23, 2024

MitchTurner left a comment

MitchTurner Oct 23, 2024

AurelienFT Oct 23, 2024

netrome Oct 24, 2024

netrome left a comment

netrome Oct 24, 2024

AurelienFT Oct 24, 2024

AurelienFT Oct 24, 2024

netrome Oct 24, 2024

rafal-ch Oct 24, 2024

AurelienFT Oct 24, 2024

netrome Oct 24, 2024

MitchTurner Oct 24, 2024

AurelienFT Oct 24, 2024

netrome Oct 24, 2024

AurelienFT Oct 24, 2024

netrome Oct 24, 2024

AurelienFT Oct 24, 2024

MitchTurner commented Oct 24, 2024

MitchTurner Oct 24, 2024

AurelienFT Oct 24, 2024

MitchTurner Oct 24, 2024

AurelienFT Oct 24, 2024

		/// If defined as -1 no limit will be applied and will use the OS limits.
		/// If not defined the system default divided by two is used.

Add new flag for maximum file descriptors in rocksdb. #2386

Are you sure you want to change the base?

Add new flag for maximum file descriptors in rocksdb. #2386

Conversation

AurelienFT commented Oct 23, 2024

Linked Issues/PRs

Description

Checklist

Before requesting review

MitchTurner left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

netrome left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

MitchTurner commented Oct 24, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment