All committers have signed the CLA.

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

✅ Deploy Preview for badger-docs canceled.

Hi, is there something I need to fix in the CL, or is this a CI issue?

This PR has been stale for 60 days and will be closed automatically in 7 days. Comment to keep it open.

This is still useful. Ping.

@pmur tests are passing now, though I'm concerned about the maintenance cost of this when we don't have an environment to test on–is that something you're able/willing to help with?

In case it's helpful, I'll point out that you can get some signal from compile-only testing via cross-compilation. For example, GOOS=aix GOARCH=ppc64 go test -c -o=/dev/null can be run on any platform you have access to, and at least check for build errors. (The -c flag tells the go command to compile the test binary but do not run it, while the -o flag is used to discard the result, since you only care about the exit code.)

I can provide some help in a limited capacity. Note, I have been using publicly available resources to implement test this (the aix golang CI instance, and the aix gcc compiler farm instance).

I hope @ayappanec might be able to provide more official help. He is part of the IBM team maintaining OSS software on aix.

@pmur Thanks for tagging me here.
We are currently in the process of creating new AIX VMs in OSU lab which we give it to various Opensource communities to enable AIX support. I will try to get a VM for this.

ping

This has been collected dust for awhile. Is this still blocking the transition of the Go AIX CI instance?

Is this still blocking the transition of the Go AIX CI instance?

Yeah, I re-checked golang/go#67299 (comment) and LUCI's cas dependency still fails to cross-compile for aix/ppc64 today:

$ GOOS=aix GOARCH=ppc64 CGO_ENABLED=0 go install go.chromium.org/luci/client/cmd/cas@latest
go: downloading go.chromium.org/luci v0.0.0-20260121191827-3391df2e9ccc
[...]
# github.com/dgraph-io/badger/v3
go/pkg/mod/github.com/dgraph-io/badger/[email protected]/dir_unix.go:62:13: undefined: unix.Flock

An alternative might be to investigate how viable or disruptive it'd be to remove the dependency on badger from its dependency tree.

I suspect there is more work to get this running fully on AIX beyond implementing this. It runs, but some unit tests seem to run incredibly slow. Though, that may be related to the only machine I have access to (gcc119 in the gcc compiler farm).

Ping

I suspect there is more work to get this running fully on AIX beyond implementing this

but some unit tests seem to run incredibly slow

I'm not keen on releasing AIX support if it's half-baked and non-performant. I do not have access to a machine running AIX.

I'm not keen on releasing AIX support if it's half-baked and non-performant. I do not have access to a machine running AIX.

I wouldn't draw that conclusion from my N=1 sample of available machines. The community machine I have access to is slow to run anything; it's a shared community machine running old (power8) hardware.

The lack of directory fsync support doesn't seem like a blocker. The documentation seems to indicate it's only best effort for OS crashes (loss of power, or otherwise).

@ayappanec could you run the tests on more representative hardware? I don't trust gcc farm machine for anything performance related.

@pmur Sure. Let me check and get back.

Is that "jemalloc" is a prerequisite for this ?

# ./test.sh
go version go1.25.5 aix/ppc64
/badger/badger /badger
github.com/dgraph-io/ristretto/v2/z
# github.com/dgraph-io/ristretto/v2/z
/go/pkg/mod/github.com/dgraph-io/ristretto/[email protected]/z/calloc_jemalloc.go:13:10: fatal error: jemalloc/jemalloc.h: No such file or directory
   13 | #include <jemalloc/jemalloc.h>
      |          ^~~~~~~~~~~~~~~~~~~~~
compilation terminated.

I ran the tests on gcc119 with something like GOMAXPROCS=8 go test -v -timeout=1h. You could try unsetting tags in test.sh for aix and running it too.

As for jemalloc, @ayappanec you may know better. I tried building the tip of jemalloc on gcc119, it failed with a nonsensical:

configure: error: Unsupported pointer size: 0

Thanks @pmur
Just ran GOMAXPROCS=8 go test -v -timeout=1h. It took around 30 minutes for completion.

Level Done
--- PASS: ExampleOpen (1.21s)
=== RUN   ExampleTxn_NewIterator
badger 2026/02/02 10:14:43 INFO: All 0 tables opened in 0s
badger 2026/02/02 10:14:43 INFO: Discard stats nextEmptySlot: 0
badger 2026/02/02 10:14:43 INFO: Set nextTxnTs to 0
badger 2026/02/02 10:14:43 INFO: Lifetime L0 stalled for: 0s
badger 2026/02/02 10:14:44 INFO: 
Level 0 [ ]: NumTables: 01. Size: 15 KiB of 0 B. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 64 MiB
Level 1 [ ]: NumTables: 00. Size: 0 B of 10 MiB. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 2.0 MiB
Level 2 [ ]: NumTables: 00. Size: 0 B of 10 MiB. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 2.0 MiB
Level 3 [ ]: NumTables: 00. Size: 0 B of 10 MiB. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 2.0 MiB
Level 4 [ ]: NumTables: 00. Size: 0 B of 10 MiB. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 2.0 MiB
Level 5 [ ]: NumTables: 00. Size: 0 B of 10 MiB. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 2.0 MiB
Level 6 [B]: NumTables: 00. Size: 0 B of 10 MiB. Score: 0.00->0.00 StaleData: 0 B Target FileSize: 2.0 MiB
Level Done
--- PASS: ExampleTxn_NewIterator (1.23s)
PASS
ok      github.com/dgraph-io/badger/v4  1777.993s

@ayappanec Can you describe the cpu and memory characteristics for the machine on which you ran that test?

# lparstat

System configuration: type=Shared mode=Uncapped smt=2 lcpu=30 mem=40960MB psize=17 ent=1.50 

%user  %sys  %wait  %idle physc %entc  lbusy  vcsw phint  %nsp  %utcyc
----- ----- ------ ------ ----- ----- ------ ----- ----- -----  ------
  0.1   0.3    0.0   99.7  0.01   0.5    3.3 3900039994     0   101   0.78 

It has 15 virtual cpus (shared mode) and smt is 2. So logical cpus is 2x15 = 30. It's a POWER8 machine
Memory is 40GB.