cnid: sqlite-default residual cleanup; dsi: raise the listen backlog to 128

[andylemin]

CNID: sqlite-default residual cleanup

SQLite has been the default CNID scheme since PR #2265 ("Build and run netatalk without Berkeley DB") made meson select sqlite > mysql > dbd among the compiled backends — and all three compile by default, so DEFAULT_CNID_SCHEME, the runtime cnid scheme default, is already "sqlite" on any default build (afpd -v: "Default CNID backend: sqlite"). What remained was residue that still said dbd:

residue	fix
docker entrypoint fallback: cnid scheme = ${AFP_CNID_BACKEND:-dbd} (two sites)	:-sqlite — an unset AFP_CNID_BACKEND now matches what a stock source build already does
~30 CI test jobs (the ea:sys/ea:ad/UAM/AFP-version/login matrix) set no AFP_CNID_BACKEND and were unintentionally testing dbd via the entrypoint fallback	they now inherit sqlite, the shipped default — only jobs that intentionally test a backend name it. The "AFP spec test (cnid:dbd)" job, which also relied on the fallback, now sets -e AFP_CNID_BACKEND=dbd explicitly
man page never stated which backend is the default	the sqlite paragraph now says "This is the default backend when compiled in."
NEWS never recorded the default change	entry added

Intentional backend coverage is unchanged: the cnid:mysql legs (Alpine + Debian) and the cnid:sqlite/perf legs set AFP_CNID_BACKEND explicitly, the VM spectest script exports dbd explicitly, and the cnid:dbd job is now pinned. Every other job had no business running dbd; this PR resolves that unintended coverage skew — the broad matrix now tests what ships.

DSI: raise the listen backlog to 128

DSI_TCPMAXPEND has been 20 since the CVS-era initial revision, with no config knob. A reconnect storm (default 200 max connections) overflows a backlog of 20: dropped SYNs and ~1s client retransmit delays. 128 matches the common SOMAXCONN floor, and the kernel silently caps the effective backlog at net.core.somaxconn, so no new option is warranted. No steady-state effect; the #ifndef guard stays for build-time overrides.

Testing

• meson test (Alpine container, CI build flags): 3/3 suites, zero failures.
• AFP spectest ea=sys and ea=ad legs run without AFP_CNID_BACKEND — i.e. on the new sqlite fallback end-to-end: both exit 0, zero failures.
• Entrypoint verification: a container pinned to AFP_CNID_BACKEND=dbd generates cnid scheme = dbd and runs the cnid daemon; an env-less container generates cnid scheme = sqlite.
• afpd -v in the built image reports sqlite as the default backend.

Outstanding issue

The netatalk master still starts cnid_metad whenever the dbd backend is merely compiled in, so an env-less (sqlite) container runs an idle cnid daemon that nothing talks to. Startup should be conditional on a volume actually using cnid scheme = dbd — tracked in #3268.

[andylemin]

@rdmark ready for review

[andylemin]

@rdmark ready for final review 🤞

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[github-actions]

📊 Performance Dashboard

Commit: e04a149a97a8c9cf2191eff84c0c8ec686c5c7df

🔥 Spectest (AFP 3.4) - FlameGraph

Netatalk Code-time: 2.7% · Runtime: 60s · Stacks: 1332

🔥 Click the preview to open the interactive flamegraph (zoom + search).

🔝 Top 10 leaf functions

Function	Samples
do_syscall_64	351708665
x64_sys_call	197353765
[libsqlite3.so.3.53.4]	127894060
_raw_spin_unlock_irqrestore	119073780
finish_task_switch.isra.0	89305335
__cp_end	78279985
afpd	77177450
srso_alias_safe_ret	52921680
kmem_cache_alloc_noprof	31973515
syscall_trace_enter	28665910

📈 Speedtest (AFP 3.4) - PerfGraph

Peak Read: 6342 MB/s (-11.4% vs hist avg 7160.9 MB/s; min 5963 / max 9526 over 27 PRs)
Peak Write: 1587 MB/s (+24.5% vs hist avg 1274.9 MB/s; min 225 / max 1863 over 27 PRs)

🔝 Throughputs per operation (vs. historical average)

Metric	Current (MB/s)	Cur Avg Δ%	Hist avg	Hist min	Hist max
Read peak mean	6342	-11.4%	7160.9	5963	9526
Read avg mean	3529	-14.9%	4147.9	3524	5393
Read avg max	4015	-12.3%	4578.9	3812	5932
Write peak mean	1587	+24.5%	1274.9	225	1863
Write avg mean	493	+18.2%	417.1	91	554
Write avg max	813	+8.9%	746.5	224	966
Copy peak mean	2535	-4.5%	2654.1	2294	3353
Copy avg mean	1467	-6.0%	1561.4	1360	1954
Copy avg max	1648	-3.3%	1704.3	1515	2087
ServerCopy peak mean	3375	-8.0%	3669.3	3055	5090
ServerCopy avg mean	1803	-6.0%	1918.3	1604	2501
ServerCopy avg max	1858	-6.7%	1992.0	1675	2658

⏱️ Lantest (AFP 3.4) - LatencyGraph

Avg total runtime: 4393 ms (+2.9% vs hist avg 4268.1 ms; min 2122 / max 5208 over 27 PRs)
Avg time per AFP op: 89 µs (+3.9% vs hist avg 85.7 µs; min 43 / max 105 over 27 PRs)

🐢 All operations (avg runtime, in test order, vs. historical average)

Metric	Current (ms)	Cur Avg Δ%	Adj Δ%	Hist avg	Hist min	Hist max
Writing one large file	39	+5.2%	—	37.1	26	47
Reading one large file	14	-9.4%	—	15.4	12	23
Creating 2000 files	497	-2.8%	-15.3%	511.2	211	831
Create 2000 dirs tree (20×9×10)	473	-7.2%	-19.7%	509.6	289	796
Open, write 1024 bytes, close 2000 files	420	+12.7%	+0.1%	372.7	195	459
Open, read 1024 bytes, close 2000 files	367	+12.6%	+0.0%	326.0	177	408
Copying 1000 files client-side (R+W)	573	+2.8%	-9.8%	557.6	267	677
Copying 2000 files server-side	480	-7.5%	-20.1%	518.9	182	727
Stat (lookup+getparams) 2000 files	252	+13.6%	+1.0%	221.9	125	282
Enumerate dir with 2000 files	11	+29.7%	+17.1%	8.48	3	14
Lock then unlock 2000 open forks	185	+13.4%	+0.9%	163.1	111	199
Deleting 2000 files	333	-3.0%	-15.5%	343.1	125	447
Byte-range lock/unlock 2000 ranges in one fork	183	+10.1%	-2.4%	166.1	115	196
Directory cache hits (20 dirs x 100 files)	115	+16.5%	+3.9%	98.7	59	122
Mixed cache operations (create/stat/enum/delete) on 500 files	222	+2.1%	-10.5%	217.4	92	265
Deep path traversal (20 levels x 100 walks)	120	+13.9%	+1.4%	105.3	61	131
Cache validation (500 files x 4 lookups)	109	+14.3%	+1.7%	95.4	58	116

Run baseline: median op-test delta +12.6%, MAD 3.9%. Adj Δ% shifts each delta by the median; standouts ≥5% in bold. A large MAD means the run did not move uniformly — read the adjusted column with caution.