rproxy now has a circuit breaker for its asks to waf (internal/rproxy/breaker.go, a field on the handler under a mutex):
5 asks with no answer in time within 5 seconds open it. Slow answers count.
While open, a request takes SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION without asking, with the same warning and waf_backend_no_answer; one request a second still asks.
The first answer in time closes it and clears the failures before it.
SPEC.md gets a "The circuit breaker" section, and the sentences saying rproxy keeps nothing between requests are rewritten in SPEC.md, README.md, docs/settings.md and the package doc.
What the diff does not make obvious:
An answer rproxy cannot pass on (status below 200, body over 1 MiB) counts as an answer in time. A visitor who leaves mid-ask counts as neither.
An answer while the breaker is closed does not clear earlier failures, so 5 failures in 5 seconds open it even among answers that came in time. Choosing to clear them instead is a one-line change.
TestTheSiteAnswersWhileWAFIsStoppedAndStartedAgain in internal/sww now polls for the ban after waf restarts: its visitor's failed asks open the breaker, which closes within a second once waf answers.
Breaker unit tests pass the time to the breaker through export_test.go, so they never sleep.
Model: opus-5-5
Closes https://git.eeqj.de/sneak/smallwebwaf/issues/392.
`rproxy` now has a circuit breaker for its asks to `waf` (`internal/rproxy/breaker.go`, a field on the handler under a mutex):
- 5 asks with no answer in time within 5 seconds open it. Slow answers count.
- While open, a request takes `SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION` without asking, with the same warning and `waf_backend_no_answer`; one request a second still asks.
- The first answer in time closes it and clears the failures before it.
`SPEC.md` gets a "The circuit breaker" section, and the sentences saying `rproxy` keeps nothing between requests are rewritten in `SPEC.md`, `README.md`, `docs/settings.md` and the package doc.
What the diff does not make obvious:
- An answer `rproxy` cannot pass on (status below 200, body over 1 MiB) counts as an answer in time. A visitor who leaves mid-ask counts as neither.
- An answer while the breaker is closed does not clear earlier failures, so 5 failures in 5 seconds open it even among answers that came in time. Choosing to clear them instead is a one-line change.
- `TestTheSiteAnswersWhileWAFIsStoppedAndStartedAgain` in `internal/sww` now polls for the ban after `waf` restarts: its visitor's failed asks open the breaker, which closes within a second once `waf` answers.
- Breaker unit tests pass the time to the breaker through `export_test.go`, so they never sleep.
Model: opus-5-5
internal/rproxy/waf.go:106-111: no test covers the two exceptions in askUnlessBreakerOpen: an ask the visitor abandons, and an answer rproxy cannot pass on, must not count as failures. The first is what stops a visitor from opening the breaker for everyone by closing connections while waf is being asked. Acceptable: handler tests in internal/rproxy/rproxy_test.go, one per exception, where five such asks in a row leave the breaker closed and the next request still asks waf.
docs/request-log.md:241: "disposition … is left out when: waf was not asked, as for the health check" is now false. A request the open breaker keeps from asking waf has disposition. Acceptable: a bullet that names only requests rproxy answers itself, such as the health check.
docs/settings.md:455-460: the touched sentence on SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION runs to 37 words and chains four conditions with commas. Acceptable: a short sentence such as "what is done with a request when waf gives no answer in time, and while the circuit breaker for asks to waf is open."
SPEC.md, "The circuit breaker": SPEC.md:9 says each requirement is stated once, but this section restates "Without an answer from waf" just above it.
Its first paragraph lists again the three cases that section gives as bullets.
Its closing paragraph (SPEC.md:2734-2735) repeats the timeout case a third time. It then says "Such a request is a bug in waf", but a request is not a bug, and SPEC.md:2697 still says "A visitor can make waf slow" as expected behaviour.
Acceptable: refer to the cases above instead of listing them again, and drop the closing paragraph. In "Without an answer from waf", say once that waf taking longer than SWW_RPROXY_WAF_BACKEND_TIMEOUT_MS to decide a request is a bug in waf.
Model: opus-5-5
Verdict: FAIL
Head: 969383894c736d9b085eb7cbfc020ab23a769578
`needs-rework`
1. `internal/rproxy/waf.go:106-111`: no test covers the two exceptions in `askUnlessBreakerOpen`: an ask the visitor abandons, and an answer `rproxy` cannot pass on, must not count as failures. The first is what stops a visitor from opening the breaker for everyone by closing connections while `waf` is being asked. Acceptable: handler tests in `internal/rproxy/rproxy_test.go`, one per exception, where five such asks in a row leave the breaker closed and the next request still asks `waf`.
2. `docs/request-log.md:241`: "`disposition` … is left out when: `waf` was not asked, as for the health check" is now false. A request the open breaker keeps from asking `waf` has `disposition`. Acceptable: a bullet that names only requests `rproxy` answers itself, such as the health check.
3. `docs/settings.md:455-460`: the touched sentence on `SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION` runs to 37 words and chains four conditions with commas. Acceptable: a short sentence such as "what is done with a request when `waf` gives no answer in time, and while the circuit breaker for asks to `waf` is open."
4. `SPEC.md`, "The circuit breaker": `SPEC.md:9` says each requirement is stated once, but this section restates "Without an answer from `waf`" just above it.
- Its first paragraph lists again the three cases that section gives as bullets.
- Its closing paragraph (`SPEC.md:2734-2735`) repeats the timeout case a third time. It then says "Such a request is a bug in `waf`", but a request is not a bug, and `SPEC.md:2697` still says "A visitor can make `waf` slow" as expected behaviour.
Acceptable: refer to the cases above instead of listing them again, and drop the closing paragraph. In "Without an answer from `waf`", say once that `waf` taking longer than `SWW_RPROXY_WAF_BACKEND_TIMEOUT_MS` to decide a request is a bug in `waf`.
Model: opus-5-5
Five asks that get no answer in time within five seconds open the
breaker. While open, a request takes
SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION without asking waf, except
one ask a second; the first answer closes it. Slow answers count as
failures. SPEC.md describes it.
The sww test that stops and restarts waf now waits for the breaker to
close before it expects the ban again.
Model: opus-5-5
Added TestVisitorsThatGoAwayWhileWAFIsAskedLeaveTheBreakerClosed and TestAnswersThatCannotBePassedOnLeaveTheBreakerClosed, the second with a deny over 1 MiB.
The bullet now reads "rproxy answered the request itself, as for the health check".
Replaced with the suggested sentence.
"The circuit breaker" refers to the cases above and drops its closing paragraph; "Without an answer from waf" states the bug once, in place of "A visitor can make waf slow".
Model: opus-5-5
Head: 7ad4bb7426e4a98cadac92768ec2907b6e13e33e
1. Added `TestVisitorsThatGoAwayWhileWAFIsAskedLeaveTheBreakerClosed` and `TestAnswersThatCannotBePassedOnLeaveTheBreakerClosed`, the second with a deny over 1 MiB.
2. The bullet now reads "`rproxy` answered the request itself, as for the health check".
3. Replaced with the suggested sentence.
4. "The circuit breaker" refers to the cases above and drops its closing paragraph; "Without an answer from `waf`" states the bug once, in place of "A visitor can make `waf` slow".
Model: opus-5-5
README.md:301, the settings table row for SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION, still says the setting applies "when waf cannot be reached or does not answer in time". It now also decides each request the open circuit breaker keeps from asking waf, including while waf is back up and the next ask has not yet closed the breaker. docs/settings.md says so, and the README row disagrees with it. Acceptable: the row says what docs/settings.md says, for example "What rproxy does with a request when waf gives no answer in time, and while the circuit breaker for asks to waf is open: allow passes it to the app, deny answers 503." Run make fmt afterwards, which re-pads the table.
Model: opus-5-5
Verdict: FAIL
Head: 7ad4bb7426e4a98cadac92768ec2907b6e13e33e
`needs-rework`
1. `README.md:301`, the settings table row for `SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION`, still says the setting applies "when `waf` cannot be reached or does not answer in time". It now also decides each request the open circuit breaker keeps from asking `waf`, including while `waf` is back up and the next ask has not yet closed the breaker. `docs/settings.md` says so, and the README row disagrees with it. Acceptable: the row says what `docs/settings.md` says, for example "What `rproxy` does with a request when `waf` gives no answer in time, and while the circuit breaker for asks to `waf` is open: `allow` passes it to the app, `deny` answers `503`." Run `make fmt` afterwards, which re-pads the table.
Model: opus-5-5
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Closes #392.
rproxynow has a circuit breaker for its asks towaf(internal/rproxy/breaker.go, a field on the handler under a mutex):SWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITIONwithout asking, with the same warning andwaf_backend_no_answer; one request a second still asks.SPEC.mdgets a "The circuit breaker" section, and the sentences sayingrproxykeeps nothing between requests are rewritten inSPEC.md,README.md,docs/settings.mdand the package doc.What the diff does not make obvious:
rproxycannot pass on (status below 200, body over 1 MiB) counts as an answer in time. A visitor who leaves mid-ask counts as neither.TestTheSiteAnswersWhileWAFIsStoppedAndStartedAgainininternal/swwnow polls for the ban afterwafrestarts: its visitor's failed asks open the breaker, which closes within a second oncewafanswers.export_test.go, so they never sleep.Model: opus-5-5
Verdict: FAIL
Head:
969383894cneeds-reworkinternal/rproxy/waf.go:106-111: no test covers the two exceptions inaskUnlessBreakerOpen: an ask the visitor abandons, and an answerrproxycannot pass on, must not count as failures. The first is what stops a visitor from opening the breaker for everyone by closing connections whilewafis being asked. Acceptable: handler tests ininternal/rproxy/rproxy_test.go, one per exception, where five such asks in a row leave the breaker closed and the next request still askswaf.docs/request-log.md:241: "disposition… is left out when:wafwas not asked, as for the health check" is now false. A request the open breaker keeps from askingwafhasdisposition. Acceptable: a bullet that names only requestsrproxyanswers itself, such as the health check.docs/settings.md:455-460: the touched sentence onSWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITIONruns to 37 words and chains four conditions with commas. Acceptable: a short sentence such as "what is done with a request whenwafgives no answer in time, and while the circuit breaker for asks towafis open."SPEC.md, "The circuit breaker":SPEC.md:9says each requirement is stated once, but this section restates "Without an answer fromwaf" just above it.SPEC.md:2734-2735) repeats the timeout case a third time. It then says "Such a request is a bug inwaf", but a request is not a bug, andSPEC.md:2697still says "A visitor can makewafslow" as expected behaviour.Acceptable: refer to the cases above instead of listing them again, and drop the closing paragraph. In "Without an answer from
waf", say once thatwaftaking longer thanSWW_RPROXY_WAF_BACKEND_TIMEOUT_MSto decide a request is a bug inwaf.Model: opus-5-5
969383894cto7ad4bb7426Head:
7ad4bb7426TestVisitorsThatGoAwayWhileWAFIsAskedLeaveTheBreakerClosedandTestAnswersThatCannotBePassedOnLeaveTheBreakerClosed, the second with a deny over 1 MiB.rproxyanswered the request itself, as for the health check".waf" states the bug once, in place of "A visitor can makewafslow".Model: opus-5-5
Verdict: FAIL
Head:
7ad4bb7426needs-reworkREADME.md:301, the settings table row forSWW_RPROXY_WAF_BACKEND_NO_ANSWER_DISPOSITION, still says the setting applies "whenwafcannot be reached or does not answer in time". It now also decides each request the open circuit breaker keeps from askingwaf, including whilewafis back up and the next ask has not yet closed the breaker.docs/settings.mdsays so, and the README row disagrees with it. Acceptable: the row says whatdocs/settings.mdsays, for example "Whatrproxydoes with a request whenwafgives no answer in time, and while the circuit breaker for asks towafis open:allowpasses it to the app,denyanswers503." Runmake fmtafterwards, which re-pads the table.Model: opus-5-5
View command line instructions
Checkout
From your project repository, check out a new branch and test the changes.