feat: halve the record buffer now the store holds the backlog

The ring is sized in records, so doubling the message cap doubled its cost. Four
records is enough to absorb a burst logged while the service task is sending; the
store, not the ring, is what holds a backlog.

  Flash      +11,980 B  (unchanged)
  RAM         +7,384 B     (-2,056)
  Log stack     +704 B  (unchanged)
  Service       +992 B  (unchanged)

Static RAM is the only figure that moves, and it is exactly four records at the cap
plus a two-byte length prefix each. Nothing else is affected: the ring bounds how
many records can be in flight, not how large one may be.

Taken with the cap rise, the pair costs 248 bytes. The ring ends up 8 bytes smaller
than it was before the cap moved — four records at 512 is less than eight at 256 —
so what is actually being paid for is the store's single record buffer, which
follows the cap and cannot be halved.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: DavidCozens

README.md

@@ -10,18 +10,18 @@ It builds on a baseline that simulates the sort of device you might be adding th
 measures itself: see [docs/baseline.md](docs/baseline.md) for what the baseline is, how the
 figures are made, and how to run it.
 
-## This stage — Larger cap
+## This stage — Smaller ring
 
-The message cap doubles to 512 bytes. Three SD-ELEMENTs had already taken a short record close to
-the old ceiling, and the two counters it carries are 32-bit — at full width the same record would
-cross it and start truncating.
+The ring holds four records instead of eight. It is sized in records, so doubling the cap doubled
+what it cost — and four is enough to absorb what gets logged while the service task is sending,
+because the store rather than the ring is what holds a backlog.
 
-The cost is almost all RAM, because the cap sizes three things at once: the ring, the store's
-record buffer, and the deepest single demand the formatter makes of whichever task calls it.
+It gives back almost everything the cap rise took. Nothing else moves: the ring bounds how many
+records can be in flight, not how large one may be.
 
 <!-- STAGE-COST:START (generated by scripts/gen-cost-table.py — do not edit by hand) -->
 
-**Cost above baseline: Flash +11,980 B, RAM +9,440 B.**
+**Cost above baseline: Flash +11,980 B, RAM +7,384 B.**
 
 <!-- STAGE-COST:END -->
 
@@ -48,6 +48,7 @@ committed as [`run-report.md`](run-report.md), and rewritten by every stage.
 | File store | records that survive a failed send, spooled to disk with a checksum at rest | +11,576 | +7,092 |
 | Origin | the device named in the record itself, not inferred from the source address | +11,972 | +7,136 |
 | Larger cap | headroom for the grown record, so full-width counters cannot push it into truncation | +11,980 | +9,440 |
+| Smaller ring | most of the cap rise given back, now the store rather than the ring holds a backlog | +11,980 | +7,384 |
 
 *Deltas are bytes above the baseline, which is itself Flash 350,308 B, RAM 111,192 B.*
 

app/syslog/Syslog.c

@@ -47,9 +47,10 @@
 #define SYSLOG_COLLECTOR_HOST "10.0.2.2"
 #define SYSLOG_COLLECTOR_PORT ((uint16_t) 5601U)
 
-/* Depth enough to absorb a burst while the sender is busy, without sizing for a
- * backlog the store is there to hold. */
-#define SYSLOG_BUFFER_RECORDS 8U
+/* Absorbs records logged while the service task is busy sending. Many devices can
+ * reduce this further: the store holds the backlog, so the ring only has to cover
+ * a burst. */
+#define SYSLOG_BUFFER_RECORDS 4U
 
 /* One "<prefix>NN.log" per block, on the volume the device already mounts. */
 #define SYSLOG_STORE_PREFIX "syslog"

measurements/buffer-halve.csv

@@ -0,0 +1,13 @@
+# buffer-halve figures (bytes) — captured by scripts/run.sh (CAPTURE=1).
+# The device reads measurements/Baseline.csv as its frozen baseline and reports current-minus-Baseline.
+flash_text,361648
+flash_data,640
+static_bss,117936
+heap_used,4440
+mbedtls_peak,21340
+mbedtls_free,11428
+lwip_mem_free,7576
+lwip_pbufs_free,13
+stack_log,824
+stack_service,1044
+stack_harness,2848

measurements/stages.tsv

@@ -23,3 +23,4 @@ time-quality	Time quality	a timestamp the collector knows how far to trust, and
 file-store	File store	records that survive a failed send, spooled to disk with a checksum at rest
 origin	Origin	the device named in the record itself, not inferred from the source address
 cap-rise	Larger cap	headroom for the grown record, so full-width counters cannot push it into truncation
+buffer-halve	Smaller ring	most of the cap rise given back, now the store rather than the ring holds a backlog

run-report.md

@@ -1,4 +1,4 @@
-# solid-syslog-example — run (cap-rise)
+# solid-syslog-example — run (buffer-halve)
 
 ## Device (self-measured)
 
@@ -12,12 +12,12 @@
 [report] key,current,baseline,used_above_baseline
 [report] flash_text,361648,349992,11656
 [report] flash_data,640,316,324
-[report] static_bss,119992,110876,9116
+[report] static_bss,117936,110876,7060
 [report] heap_used,4440,4440,0
-[report] mbedtls_peak,21332,21332,0
-[report] mbedtls_free,11436,11436,0
+[report] mbedtls_peak,21340,21332,8
+[report] mbedtls_free,11428,11436,-8
 [report] lwip_mem_free,7576,7576,0
-[report] lwip_pbufs_free,14,14,0
+[report] lwip_pbufs_free,13,14,-1
 [report] stack_log,824,120,704
 [report] stack_service,1044,52,992
 [report] stack_harness,2848,2840,8
@@ -29,7 +29,7 @@
 
 ```text
    text	   data	    bss	    dec	    hex	filename
- 361640	    648	 119992	 482280	  75be8	/w/build/baseline-cross/baseline.elf
+ 361640	    648	 117936	 480224	  753e0	/w/build/baseline-cross/baseline.elf
 ```
 
 ## Listeners (proved before the device ran)
@@ -47,21 +47,21 @@
 ## Collector (syslog-ng) received
 
 ```text
-wire   <134>1 2026-07-29T08:25:48.280000Z 10.0.2.15 solid-syslog-example - BOOT [meta sequenceId="1" sysUpTime="228"][timeQuality tzKnown="1" isSynced="0"][origin software="solid-syslog-example" swVersion="0.1.0" enterpriseId="32473"] device started
-parsed PRIORITY=134 TIMESTAMP=2026-07-29T08:25:48+00:00 HOSTNAME=10.0.2.15 APP_NAME=solid-syslog-example PROCID= MSGID=BOOT STRUCTURED_DATA=[meta sequenceId="1" sysUpTime="228"][timeQuality tzKnown="1" isSynced="0"][origin software="solid-syslog-example" swVersion="0.1.0" enterpriseId="32473"] MSG=device started
+wire   <134>1 2026-07-29T08:29:37.410000Z 10.0.2.15 solid-syslog-example - BOOT [meta sequenceId="1" sysUpTime="241"][timeQuality tzKnown="1" isSynced="0"][origin software="solid-syslog-example" swVersion="0.1.0" enterpriseId="32473"] device started
+parsed PRIORITY=134 TIMESTAMP=2026-07-29T08:29:37+00:00 HOSTNAME=10.0.2.15 APP_NAME=solid-syslog-example PROCID= MSGID=BOOT STRUCTURED_DATA=[meta sequenceId="1" sysUpTime="241"][timeQuality tzKnown="1" isSynced="0"][origin software="solid-syslog-example" swVersion="0.1.0" enterpriseId="32473"] MSG=device started
 ```
 
-## Self-check (vs measurements/cap-rise.csv)
+## Self-check (vs measurements/buffer-halve.csv)
 
 ```text
   OK    flash_text: 361648 (expected 361648, Δ0)
   OK    flash_data: 640 (expected 640, Δ0)
-  OK    static_bss: 119992 (expected 119992, Δ0)
+  OK    static_bss: 117936 (expected 117936, Δ0)
   OK    heap_used: 4440 (expected 4440, Δ0)
-  OK    mbedtls_peak: 21332 (expected 21332, Δ0)
-  OK    mbedtls_free: 11436 (expected 11436, Δ0)
+  OK    mbedtls_peak: 21340 (expected 21340, Δ0)
+  OK    mbedtls_free: 11428 (expected 11428, Δ0)
   OK    lwip_mem_free: 7576 (expected 7576, Δ0)
-  OK    lwip_pbufs_free: 14 (expected 14, Δ0)
+  OK    lwip_pbufs_free: 13 (expected 13, Δ0)
   OK    stack_log: 824 (expected 824, Δ0)
   OK    stack_service: 1044 (expected 1044, Δ0)
   OK    stack_harness: 2848 (expected 2848, Δ0)