Replace node_size with memory_limit by dangell7 · Pull Request #7965 · XRPLF/rippled

dangell7 · 2026-08-06T16:12:10Z

High Level Overview of Change

Replaces [node_size] with [memory_limit]: the memory budget, in gigabytes, that the server sizes its caches within. When it is not set, the budget defaults to detected physical RAM, and on Linux the detection respects the container or service cgroup limit. Setting 0 selects minimal sizes with no enforcement. [node_size] still parses as a deprecated alias (tiny/small/medium/large/huge map to 4/8/32/64/128 GB) and warns at startup, so existing configs keep working.

Context of Change

node_size has served for a long time, but its five presets are entry counts, which are hard to reason about in bytes, and the targets are advisory. During capacity testing our 64 GB nodes ran out of memory at about 2.5M accounts on the "huge" preset, while the same nodes on "large" carried the same load at 2.3 GiB RSS. That test is the calibration for this change: a 64 GB budget produces the "large" sizing that held up, and 128 GB produces the old "huge". The model is the one Redis uses with maxmemory, a byte budget the process keeps.

The change:

Every sized value derives from the budget. The tree cache gets half of it at an estimated 8 KiB per entry, the db caches scale between their old floors and ceilings, and the time and policy values become fixed defaults with their own override sections (tree_cache_age, ledger_cache_age, ledger_fetch_size).
The tree cache target is enforced. TaggedCache gains an opt-in hard cap that demotes the approximately oldest entry whenever growth would exceed it, on fresh inserts and weak-to-strong revivals, with bounded work per call. The sweep remains the primary evictor, and saturation is logged.
Thread counts derive from the core count alone; [workers] and [io_workers] still override.
[node_db] cache_size and cache_age now default to the documented 16384 records and 5 minutes. They previously inherited the tree cache values, and the age was read in the wrong unit, which gave a "huge" node a second 8.4M-entry cache with 15-hour retention.
Admin server_info reports memory_limit in GB instead of node_size (API-CHANGELOG updated), and the unused SizedItem::OpenFinalLimit and RamSizeGb entries are removed.

API Impact

libxrpl change: new defaulted TaggedCache constructor parameter; Config::getValueFor loses its node-size override parameter; two dead SizedItem entries removed.
Public API: admin server_info drops node_size, adds memory_limit (see API-CHANGELOG.md).

Test Plan

Config tests cover parsing (gigabytes, 0, garbage and values over 1024 rejected), the budget-derived values, standalone behavior, and the alias forms (names, the legacy 0-4 index, and [memory_limit] winning when both are set). GTest covers the hard cap holding through a 1000-insert burst with the sweep unable to fire, and cap 0 disabling it. The full unit test suite passes.

Future Tasks

Extend memory_limit to the remaining caches #7966 extends the budget to the remaining caches with byte charging and a memory pressure breaker.
Remove the [node_size] alias at the end of the deprecation period.

xrplf-ai-reviewer

Two budget-aware cache sizing issues flagged inline.

xrplf-ai-reviewer · 2026-08-06T16:17:41Z

        }

-        if (!section.exists(Keys::kFilterBits) && (config.nodeSize >= 2))
+        if (!section.exists(Keys::kFilterBits))


Bloom filter unconditionally enabled for minimal-budget configs. Re-gate on budget:

Suggested change

if (!section.exists(Keys::kFilterBits))

if (!section.exists(Keys::kFilterBits) && config.cacheMemoryBudget() != 0)

✅ Auto-resolved: the latest review no longer flags this issue, so it appears to have been addressed. If that is not correct, reopen this thread and it will be re-checked on the next review.

xrplf-ai-reviewer · 2026-08-06T16:17:41Z

+            return std::clamp(gb * 2, 4, 128);
+        case SizedItem::BurstSize:
+            return std::clamp(gb, 4, 48);
+        case SizedItem::AccountIdCacheSize:


AccountIdCacheSize hardcoded to 300007 ignores new budget. Scale like other caches:

Suggested change

case SizedItem::AccountIdCacheSize:

return std::clamp(gb * 2344, 20047, 300007);

✅ Auto-resolved: the latest review no longer flags this issue, so it appears to have been addressed. If that is not correct, reopen this thread and it will be re-checked on the next review.

xrplf-ai-reviewer

AccountIdCacheSize scaling inconsistency—see inline.

xrplf-ai-reviewer · 2026-08-06T16:30:18Z

+            return std::clamp(gb * 2, 4, 128);
+        case SizedItem::BurstSize:
+            return std::clamp(gb, 4, 48);
+        case SizedItem::AccountIdCacheSize:


AccountIdCacheSize hardcoded to 300007 while other db caches scale with budget—document why or scale it:

case SizedItem::AccountIdCacheSize: // Fixed regardless of budget: account ID cache is not // memory-significant and uses a documented policy value. return 300007;

✅ Auto-resolved: the latest review no longer flags this issue, so it appears to have been addressed. If that is not correct, reopen this thread and it will be re-checked on the next review.

Done, kept it fixed with the comment. It's ~22 MB at most and the value stays prime for hash distribution. resolved: f80e72c6f7

✅ Auto-resolved: the latest review no longer flags this issue, so it appears to have been addressed. If that is not correct, reopen this thread and it will be re-checked on the next review.

xrplf-ai-reviewer

Bloom filter config and log-level issues flagged inline.

xrplf-ai-reviewer · 2026-08-06T16:43:47Z

        }

-        if (!section.exists(Keys::kFilterBits) && (config.nodeSize >= 2))
+        if (!section.exists(Keys::kFilterBits))


Bloom filter now enabled unconditionally for all configs, increasing memory in minimal deployments. Gate on cacheMemoryBudget():

Suggested change

if (!section.exists(Keys::kFilterBits))

if (!section.exists(Keys::kFilterBits) && config.cacheMemoryBudget() != 0)

Done, gated on a nonzero budget. resolved: f80e72c6f7

✅ Auto-resolved: the latest review no longer flags this issue, so it appears to have been addressed. If that is not correct, reopen this thread and it will be re-checked on the next review.

xrplf-ai-reviewer · 2026-08-06T16:43:47Z

+                  : 0))
 {
+    auto const budget = app.config().cacheMemoryBudget();
+    JLOG(j_.warn()) << "TreeNodeCache sizing: target="


Normal startup log at warn level will trigger monitoring alerts. Use j_.info() for routine messages:

Suggested change

JLOG(j_.warn()) << "TreeNodeCache sizing: target="

JLOG(j_.info()) << "TreeNodeCache sizing: target="

Done, moved to info. resolved: f80e72c6f7

✅ Auto-resolved: the latest review no longer flags this issue, so it appears to have been addressed. If that is not correct, reopen this thread and it will be re-checked on the next review.

codecov · 2026-08-06T18:16:06Z

Codecov Report

❌ Patch coverage is 71.92118% with 57 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/xrpld/core/detail/Config.cpp	66.1%	43 Missing ⚠️
include/xrpl/basics/TaggedCache.ipp	80.4%	11 Missing ⚠️
src/xrpld/app/main/Application.cpp	50.0%	2 Missing ⚠️
src/xrpld/app/misc/SHAMapStoreImp.cpp	80.0%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

xrplf-ai-reviewer

Bloom filter scope expanded — see inline.

xrplf-ai-reviewer · 2026-08-06T19:14:40Z

        }

-        if (!section.exists(Keys::kFilterBits) && (config.nodeSize >= 2))
+        if (!section.exists(Keys::kFilterBits) && config.cacheMemoryBudget() != 0)


Bloom filter bits enabled for all budgets (was medium+ tiers only). Likely unintended scope expansion for a memory-reduction PR — gate on threshold:

Suggested change

if (!section.exists(Keys::kFilterBits) && config.cacheMemoryBudget() != 0)

if (!section.exists(Keys::kFilterBits) && config.cacheMemoryBudget() >= (16ull << 30))

feat: replace node_size with memory_limit

45ed9e4

xrplf-ai-reviewer Bot reviewed Aug 6, 2026

View reviewed changes

dangell7 added 2 commits August 6, 2026 12:18

feat: resolve the process cgroup path for nested memory limits

c65e453

feat: add overrides for fixed cache policy values

a40b884

xrplf-ai-reviewer Bot reviewed Aug 6, 2026

View reviewed changes

fix: harden cgroup detection and rename ledger_fetch_size

c16f18f

xrplf-ai-reviewer Bot reviewed Aug 6, 2026

View reviewed changes

dangell7 mentioned this pull request Aug 6, 2026

Extend memory_limit to the remaining caches #7966

Draft

1 task

fix: address clang-tidy findings and review feedback

f80e72c

xrplf-ai-reviewer Bot reviewed Aug 6, 2026

View reviewed changes

dangell7 marked this pull request as draft August 7, 2026 05:31

	if (!section.exists(Keys::kFilterBits))
	if (!section.exists(Keys::kFilterBits) && config.cacheMemoryBudget() != 0)

	case SizedItem::AccountIdCacheSize:
	return std::clamp(gb * 2344, 20047, 300007);

	JLOG(j_.warn()) << "TreeNodeCache sizing: target="
	JLOG(j_.info()) << "TreeNodeCache sizing: target="

Uh oh!

Conversation

dangell7 commented Aug 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

High Level Overview of Change

Context of Change

API Impact

Test Plan

Future Tasks

Uh oh!

xrplf-ai-reviewer Bot left a comment

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot left a comment

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

dangell7 Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot left a comment

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

dangell7 Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

dangell7 Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

codecov Bot commented Aug 6, 2026

Codecov Report

Uh oh!

xrplf-ai-reviewer Bot left a comment

Choose a reason for hiding this comment

Uh oh!

xrplf-ai-reviewer Bot Aug 6, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

dangell7 commented Aug 6, 2026 •

edited

Loading