starknet_committer: add mdbx storage

[yoavGrs]

No description provided.

[reviewable-StarkWare]

This change is 

[yoavGrs]

• starknet_committer: add mdbx storage #8917 : 2 dependent PRs (#9276 , #9505 ) 👈 (View in Graphite)
• committer: add cli for committer benchmark #9345
• committer_storage_benchmark

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

Artifacts upload workflows:

• Started at 2025-08-28T12:47:42Z
• Started at 2025-09-01T07:22:26Z
• Started at 2025-09-07T06:57:27Z
• Started at 2025-09-15T08:50:03Z
• Started at 2025-09-15T10:13:20Z
• Started at 2025-09-15T13:16:17Z
• Started at 2025-09-16T13:05:52Z
• Started at 2025-09-16T13:05:52Z
• Started at 2025-09-16T13:24:33Z
• Started at 2025-09-17T09:45:29Z
• Started at 2025-09-17T10:55:00Z
• Started at 2025-09-17T10:55:06Z
• Started at 2025-09-17T11:29:27Z
• Started at 2025-09-17T11:37:40Z
• Started at 2025-09-17T11:37:56Z
• Started at 2025-09-17T14:45:32Z
• Started at 2025-09-18T07:51:04Z
• Started at 2025-10-05T12:07:47Z
• Started at 2025-10-05T12:18:55Z
• Started at 2025-10-05T13:15:48Z

[github-actions]

Benchmark movements: No major performance changes detected.

[dorimedini-starkware]

@dorimedini-starkware reviewed 3 of 6 files at r4, 3 of 3 files at r5, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @TzahiTaub)

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 21 at r5 (raw file):

                // LIFO policy for recycling a Garbage Collection items should be faster when using
                // disks with write-back cache.
                liforeclaim: true,

what's this...?

Code quote:

                // LIFO policy for recycling a Garbage Collection items should be faster when using
                // disks with write-back cache.
                liforeclaim: true,

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 43 at r5 (raw file):

        let table = txn.open_table(None)?;
        let prev_val = txn.get(&table, &key.0)?.map(DbValue);
        txn.put(&table, key.0, value.0, WriteFlags::UPSERT)?;

what's this?

Code quote:

WriteFlags::UPSERT

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 54 at r5 (raw file):

        for key in keys {
            res.push(txn.get(&table, &key.0)?.map(DbValue));
        }

libmdbx doesn't have a faster implementation of multi-read?

Code quote:

        for key in keys {
            res.push(txn.get(&table, &key.0)?.map(DbValue));
        }

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 63 at r5 (raw file):

        for (key, value) in key_to_value {
            txn.put(&table, key.0, value.0, WriteFlags::UPSERT)?;
        }

same Q (no faster impl?)

Code quote:

        for (key, value) in key_to_value {
            txn.put(&table, key.0, value.0, WriteFlags::UPSERT)?;
        }

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 75 at r5 (raw file):

            txn.del(&table, &key.0, None)?;
        }
        txn.commit()?;

is this required if prev_val is None?

Code quote:

txn.commit()?;

[dorimedini-starkware]

@dorimedini-starkware reviewed 16 of 16 files at r7, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @TzahiTaub)

[TzahiTaub]

@TzahiTaub reviewed 1 of 6 files at r4, 1 of 3 files at r5.
Reviewable status: 3 of 20 files reviewed, all discussions resolved (waiting on @dorimedini-starkware and @yoavGrs)

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 21 at r5 (raw file):

Previously, dorimedini-starkware wrote…

what's this...?

Added a TODO to go deeper into this later.

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 43 at r5 (raw file):

Previously, dorimedini-starkware wrote…

what's this?

UPdate if key exists, inSERT otherwise.

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 54 at r5 (raw file):

Previously, dorimedini-starkware wrote…

libmdbx doesn't have a faster implementation of multi-read?

Don't see one.

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 63 at r5 (raw file):

Previously, dorimedini-starkware wrote…

same Q (no faster impl?)

Don't see one.

[TzahiTaub]

Reviewable status: 3 of 20 files reviewed, all discussions resolved (waiting on @dorimedini-starkware and @yoavGrs)

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 75 at r5 (raw file):

Previously, dorimedini-starkware wrote…

is this required if prev_val is None?

Nope

[dorimedini-starkware]

@dorimedini-starkware reviewed 2 of 17 files at r8, 1 of 1 files at r9, all commit messages.
Reviewable status: 6 of 20 files reviewed, all discussions resolved (waiting on @yoavGrs)

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 22 at r9 (raw file):

// TODO(tzahi): geometry and related definitions are taken from apollo_storage. Check if there are
// better configurations for the committer and consider moving boh crates mdbx code to a common
// location.

move this TODO to MdbxStorage::open please

Suggestion:

// TODO(tzahi): geometry, liforeclaim and related definitions are taken from apollo_storage. Check if there are
// better configurations for the committer and consider moving boh crates mdbx code to a common
// location.

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 26 at r9 (raw file):

// Size in bytes.
const MDBX_MIN_PAGESIZE: usize = 256;
const MDBX_MAX_PAGESIZE: usize = 65536; // 64KB

non-blocking, but if it works - it looks nicer

Suggestion:

const MDBX_MIN_PAGESIZE: usize = 1 << 8;
const MDBX_MAX_PAGESIZE: usize = 1 << 16; // 64KB

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 34 at r9 (raw file):

    if !page_size.is_power_of_two() {
        page_size = page_size.next_power_of_two() / 2;
    }

any reason to clamp again after this?
is it possible that MDBX_MIN_PAGESIZE will not be a power of two?

Code quote:

    // Page size must be power of two.
    if !page_size.is_power_of_two() {
        page_size = page_size.next_power_of_two() / 2;
    }

[TzahiTaub]

Reviewable status: 3 of 20 files reviewed, all discussions resolved (waiting on @dorimedini-starkware and @yoavGrs)

---

crates/starknet_patricia_storage/src/mdbx_storage.rs line 34 at r9 (raw file):

Previously, dorimedini-starkware wrote…

any reason to clamp again after this?
is it possible that MDBX_MIN_PAGESIZE will not be a power of two?

This is a copy of the code in apollo_storage. I don't think it can be something else, but I wouldn't want to diverge from or change the code in apollo_storage in this PR.
And see (old answers here): https://stackoverflow.com/questions/40518454/is-it-safe-to-assume-memory-page-size-is-a-power-of-two
Seems that it would always be a power of 2 unless someone is making an exercise in a CS course.

[dorimedini-starkware]

@dorimedini-starkware reviewed 3 of 3 files at r10, all commit messages.
Reviewable status: 6 of 20 files reviewed, all discussions resolved (waiting on @yoavGrs)

[dorimedini-starkware]

@dorimedini-starkware reviewed 14 of 17 files at r8.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @yoavGrs)

[TzahiTaub]

@TzahiTaub reviewed 1 of 6 files at r4, 3 of 3 files at r10, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @yoavGrs)