runtime: reserve 16GB of heap address space on 64-bit unix - #5644
Open
0pcom wants to merge 1 commit into
Open
Conversation
0pcom
force-pushed
the
unix-heap-reserve
branch
from
September 2, 2026 08:14
68058ae to
ae3954c
Compare
Contributor
Author
|
Fixed the CI failures: the 16GB literal did not compile on 32-bit targets (ARM/MIPS) — Go type-checks the dead branch too. The reservation size is now a constant derived from TargetBits ( |
allocateHeap capped the heap at 1GB for all targets, so with -gc=conservative or -gc=precise any single allocation approaching 1GB (for example a scrypt key-derivation buffer with N=1<<20, r=8) failed with out of memory regardless of available system RAM. Reserve 16GB of virtual address space on 64-bit targets instead. The mmap is a reservation, not a commitment: pages cost physical memory only when first touched, and the existing halve-on-failure loop still adapts when the map is refused. 32-bit targets keep the 1GB cap; the size is derived from TargetBits as a shifted constant, since a plain 16GB literal does not compile on 32-bit targets even in a dead branch. This is the direction the growHeap comment already points at: "If we run out of memory, we should consider increasing heapMaxSize on 64-bit systems." With this, the practical limit under the blocks GC on 64-bit hosts becomes actual system memory, matching the boehm default and big Go; true exhaustion surfaces as the OS's memory pressure handling rather than a fatal error at an arbitrary 1GB.
0pcom
force-pushed
the
unix-heap-reserve
branch
from
September 2, 2026 09:05
ae3954c to
5572fc4
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
allocateHeap caps the heap at 1GB for all targets, so with
-gc=conservativeor-gc=preciseany single allocation approaching 1GB fails with out of memory regardless of available system RAM. The real case that hit this: a scrypt key-derivation buffer (N=1<<20, r=8 → 1GiB) in the skycoin wallet, on a machine with 16GB free.Reserve 16GB of virtual address space on 64-bit targets instead. The mmap is a reservation, not a commitment: pages cost physical memory only when first touched, and the existing halve-on-failure loop still adapts when the map is refused. 32-bit targets keep the 1GB cap unchanged. This is the direction the
growHeapcomment already points at: "If we run out of memory, we should consider increasing heapMaxSize on 64-bit systems."Verified on linux/amd64 with
-gc=conservative:make([]byte, 1536<<20)→fatal error: out of memory(15GB RAM free)One behavioral note, stated for the record: under the blocks GC on 64-bit hosts the practical limit becomes actual system memory — matching the boehm default and big Go — so true exhaustion surfaces as OS memory pressure rather than a fatal error at an arbitrary 1GB.
🤖 Generated with Claude Code
https://claude.ai/code/session_01C3X3gvq8ZggMRvzVzWm66i