Repeat the REPORT.md "Запуск из чистого clone" sequence from a real git
clone of the final strict-id-validation code on the default ports: env
setup, stand, no-cache build, check, validate, smoke, instances on two
Python versions, persistence, failover of each replica and kill, own
tests, fault checks and the isolated run. All steps exit 0. The clone ran
as compose project mtusi-lab3-cleancheck while the working stand was
only stopped, because removing its volume was refused by the agent's
auto mode; the previous checks ran as mtusi-lab3.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its delta specs were synced into api-conventions, chats and pagination
during apply and match the main specs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The docstring said every field is checked for the server's notation;
only i is, t is checked for type, and unknown fields are ignored. Also
note in tasks.md that the re-encoding check of 3.3 was replaced in 7.2
and fix the rules table. Behaviour of 8fe467a is unchanged; lint and
make test_isolated passed again (138 + 4).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Save the red run on c5d5a1a, the checks of the first fix (628b5f2) and
of the cursor field checks that replaced re-encoding (8fe467a): isolated
tests, rolling update, smoke and own tests through the load balancer
with no 5xx, fault checks, and a clean clone in a temporary compose
project on other ports. Update REPORT.md with the new evidence row, the
behaviour change for non-canonical UUIDs and the current test counts,
and AGENTS.md with the change and the code map.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
628b5f2 accepted a cursor only if re-encoding the parsed key gave the
same string. That closed the format both ways: any encoding change, even
a compatible one (an optional field, key order, to_rfc3339 shared with
API responses), would turn cursors issued by a replica of the other
version into 400 during make rolling_update, which the tests cannot see.
decode_cursor now checks the fields one by one: v must be exactly the
integer 1, t and i must be strings, i must be in the server's notation
(lowercase 8-4-4-4-12). Unknown fields are ignored, so a tampered cursor
is still a 400 while the format stays open for compatible changes. The
rule records how to change the format: a new CURSOR_VERSION, with both
versions accepted while replicas are replaced.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
decode_cursor checked only for KeyError, TypeError and ValueError, so a
server-issued cursor whose item id was replaced with [], {}, 123 or
false crashed UUID() with AttributeError and answered 500. It now checks
the types of "t" and "i" before parsing and accepts only the exact
string the server would issue for the parsed key: re-encoding must give
the received cursor, so v=true, an extra field or the id in another
notation is a 400 instead of a page.
pydantic's UUID also accepts ids without hyphens, with the urn:uuid:
prefix and in braces, wider than the contract's format: uuid. chat_id in
the path of the three chat endpoints and member_ids elements now use
UuidIn, which requires the 8-4-4-4-12 form in any case. Response models
and the OpenAPI schema are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add black-box tests for the strict-id-validation change: a server-issued
cursor with one field changed (non-string or non-canonical item id,
v=true, extra field) on both lists, member_ids with values of other JSON
types, a non-array member_ids and other UUID notations, and chat_id in
the path of the three chat endpoints. Sync the change's delta specs into
api-conventions, chats and pagination so the coverage check knows the
new scenarios.
The cursor and UUID-notation tests fail on the current code by design;
the fix follows in the next commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The change moves to openspec/changes/archive/2026-10-13-fix-known-issues.
Its delta requirements join the main specs: graceful-shutdown "Остановка
во время старта", load-benchmark "События Docker профилей отказа и
замены", student-tests "Доказательства исключений сверки в репозитории"
and "PostgreSQL стенда отказов без аварийных перезапусков". AGENTS and
REPORT now say all eight lab 3 changes are archived and none is active.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Evidence section "Правки по verify": stop-at-start under system Python
3.9.6, cache-check with and without --require-test-keys, auth-storage,
the postgres log read failure path, student tests, failure checks and
the quick failover and rollout profiles on f02694e. The REPORT header
lists what changed after the clean clone commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
REPORT: 133 and 129 + 4 in the clean clone section, the postgres log
summary of failure_checks, the isolated count on the final code.
AGENTS: do not copy lab 3 evidence, cache-check --require-test-keys.
script-failures: the closed KNOWN_ISSUES entry points to decision 7.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stop_stream terminates the events CLI, waits 10 s and kills it after
that. The rollout profile starts k6 inside the try, so a k6 start
failure no longer leaves the project events stream running.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
docker compose logs runs with check=False; a non-zero exit prints
"postgres log unavailable" and fails the run instead of being reported
as a setup failure.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cache-check gains --require-test-keys: without keys of the test database
it fails instead of passing, and a plain PASS now says the temporary
database was not checked. The coverage exclusion names the flag.
auth-storage logs in only the first user, as the design says.
stop-at-start installs its signal handlers right before the try that
restores them and kills docker logs -f when reading is interrupted.
The student-tests rule warns that copied evidence keeps the check green.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The *.log rule kept LB, rollout script and 5xx logs of committed
labels out of git while REPORT.md cites them. The 59 logs (556 KB)
hold no tokens, passwords or host paths; quick-* labels stay ignored.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Report the failover ranges strictly over phases 1-5 with phase 06
apart, add per-window memory peaks to the failover table, describe
which intervals the window maximum covers, and note the per-second
rows and per-request zeros of the taken labels fixed in f34cb9a.
Record the design findings, the raw logs ignored by git as a known
issue, and the checks after the fixes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The samples.csv row of an interval that touches the start of a freeze
now gets the window of its unfrozen part, as the window aggregates do
since the thaw correction. CPU per successful request is no longer
computed for a container absent from the window. The definitions in
the summary describe which intervals the window maximum covers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A sample interval that crosses docker pause or unpause carried the CPU
spent before the pause and the burst after unpause into the failover
window by proportional split. Use the pause and unpause moments from
Docker events so that interval CPU lands outside the frozen span.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Run bench/sampler.sh in a new scan mode (cgroup walk, no network, no
psql or LB polling) for the whole load of the failover and rollout
profiles. bench/resources.py turns the samples into per-window CPU,
throttling and memory of every component and the VM, with
samples.csv, resources.csv and summary sections; CPU per successful
request only in steady windows. Check the peak CPU budget before the
stand starts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fix the findings of the final REPORT.md review: prerequisites, the
defence scenario, swap sources, the single explanation of the R* drop
at PostgreSQL 0.5 CPU, known defects in the limitations, a dead link,
stale counts and references. Remove the device name from evidence.
Remove the lab plan now that every change is archived: its decisions
live in REPORT.md and the archived designs, live references point
there, and AGENTS.md drops the transitional section.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
datetime.UTC exists only since Python 3.11, so make instances failed on
the macOS system python3 listed in the report prerequisites. Use
timezone.utc and postponed annotations instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fill the remaining report sections: clean clone run with versions and
prerequisites, lab 2 carry-over, transaction boundaries, teacher
agreements, skeleton variable mapping, rubric cross-check, experiment
overview, limitations summary, sources and AI usage. Record the clean
clone run of every documented command on default ports.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sync load-benchmark (series comparison table, track 1 before/after
rules, db4 series) and service-configuration (default pool 10 + 0)
into the main specs, move the change to the archive and drop it from
the transitional state in AGENTS.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fix the two-step threshold (x1.118), CPU-per-request spread and
PostgreSQL ranges in REPORT.md and design, record the LB side check
(lb_chkdown, error_codes) in evidence, add the db4 series to the
load-benchmark delta, refresh AGENTS.md, bench.md, PLAN.md and
proposal, and log the swapouts_delta passport defect.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Configs may carry a fifth element, the PostgreSQL CPU limit, which goes
into the config key (-pg<CPU>), the budget and the cell stand. Series db:
1, 2, 4 replicas at the tight limit and a 4 x 1 control at 2 CPU; label
summaries show PostgreSQL CPU at R* against its limit and throttling.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Close the 53300 known issue, restate H1 and saturation conclusions
against the pre-registered criteria and grid resolution, name the
unresolved host mechanism, attribute overload 503 to the pool timeout,
and plan a fresh A/A for scaling-improvement.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
plots.py summary gains per-request CPU by component, PostgreSQL waits and
pool, H1b throttling and memory peaks; compare.py and plots.py headers
name the spec, change and host; errors logs match 5xx by status field and
keep the exception tail; series labels record sampler limits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SQLSTATE 53300 (too_many_connections) joins UNAVAILABLE_SQLSTATES, so a
refused pool connection gives 503 unavailable instead of 500. New fault
check fills max_connections through the failure stand hook and expects
503 in the contract format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Series a/b/pool/ab/quick in bench/run.py: per-cell fresh stand with the
configuration's replicas and max_connections, open-model mixed profile
(k6/mixed.js), fixed levels and R* search by SLO, a sampler container
for cgroup, VM, HAProxy and PostgreSQL metrics, resume and a stand
lock. Pure helpers in bench/scaling.py, SVG plots in bench/plots.py,
series mode for bench_compare and bench_ab. Hypotheses H1-H3 go to
REPORT.md before the first series.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Track 1 "cost of scaling": open-model series A (fixed 4 CPU budget),
B (1-6 replicas), pool border (max_connections 100 vs 50) and A/A,
sampler of cgroup/VM/LB/PostgreSQL metrics, generator limits, series
mode for bench_compare and bench_ab. Hypotheses H1-H3 are recorded in
the design before any series run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sync the rolling-update delta specs into db-migrations,
graceful-shutdown, load-balancing and load-benchmark, move the change to
the archive, mark it in PLAN.md and update the transitional state and
code map in AGENTS.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Re-run the compatibility walk-through, the script failure cases with a
real Ctrl+C through a pseudo-terminal, the stand lifecycle after
replacement and the quick rollout and failover self-checks on 84161c0
with a clean tree; update the report rows and limitations.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With lock_timeout equal to statement_timeout (5 s) every request queued
behind a migration waiting for its lock failed with 503 together with
the migration. A 2 s lock timeout lets the queue proceed once the
migration gives up.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reset the new replica ID before entering the starting state and let the
cleanup find a replica created by an interrupted compose scale; refuse
to start while LB slots keep a manual drain or maint state; log the exit
code of each stopped old replica and print full LB commands for manual
cleanup. Document the migrate step and the rollback tag limitation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A one-off run --rm left the migrate container of the previous image in
the stand; scripts/restart.sh then started it as a dependency of api and
the old migrations failed on the newer schema, so test-persistence broke.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Old image passes smoke and student tests on the new schema, rolling
update, rollback and roll forward under tests/load.py give zero failed
requests; the run_docker known issue is not reproduced and is closed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scripts/rolling-update.sh (make rolling_update) replaces API replicas one
by one: build, migrations once, new replica UP in HAProxy, old replica
drained and put into maintenance through the LB runtime API on loopback,
then stopped and removed; the slot is reset once DNS drops the address.
Expand migration adds messages.instance_id with a lock timeout. The bench
rollout profile runs the same script under constant load, starting from
the image of an older commit built with git archive.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Delta specs synced into graceful-shutdown, load-balancing, load-benchmark
and service-configuration; the change moves to
openspec/changes/archive/2026-10-11-graceful-failover/. PLAN.md marks the
change archived and AGENTS.md notes what was handed to rolling-update.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The verify fix in bench/compare.py refuses a label measured without the
load balancer; the load-benchmark delta now modifies the label comparison
requirement accordingly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The rejected LB pool variant is re-measured on committed code with a
temporary second HAProxy differing in http-reuse only: the replica, not
the LB, closed the last idle connection after 60.4 s. Exploratory
never/safe SIGTERM numbers are separated from evidence. hang-key cleanup
is shown under Ctrl+C and SIGTERM, bench_compare refuses a label without
the LB. Drain before SIGTERM and the run_docker recreate issue are handed
to rolling-update in PLAN.md and KNOWN_ISSUES.md; design gains decision 10.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
hang-key names its trigger per run, turns SIGTERM and SIGHUP into an
interrupt so the trigger is always dropped and the replica unpaused, and
prints the cleanup SQL if dropping fails. idle-conns reports who closed
the idle connections and can read the pool of another LB container, so a
variant LB config can be checked against the same replicas. bench_compare
refuses a label measured without the load balancer. Stale numbers removed
from the haproxy.cfg comment.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Failover profile label baseline from 459f84c on a checked host: stop and
kill of each replica under 200 req/s give 0 errors and LB exclusion in
0.79-1.81 s (bound 2.5 s), pause gives LB exclusion in 5.83-6.00 s
(bound 6.5 s) with 504s for half the requests and recovery in 6 s;
reconciliation of retried writes matches. Evidence for detection,
keep-alive (LB closes idle replica connections at 10.1 s) and the hung
replica key lock, regression checks, timeout chain and level 4 rows in
REPORT.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
HAProxy checks get fastinter 500ms and timeout check 2.5s (hung replica
detected within 6.5 s), http-reuse never with uvicorn keep-alive 60 s so
the LB closes idle replica connections, and app connections set
idle_in_transaction_session_timeout so a frozen replica releases its
idempotency key. The bench stand runs behind HAProxy with N replicas and
gains the failover profile (make bench_failover / bench_failover_quick):
constant-rate writes with a key and reads while each replica is stopped,
killed or paused, with per-second timeline, detection and recovery time
and write reconciliation. Driver commands idle-conns and hang-key check
the keep-alive and hung-transaction decisions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Main specs were already in sync with the deltas (sync re-run, no
changes). PLAN.md marks stage 2 archived and points graceful-failover
at the corrected detection formula; AGENTS.md notes what of the timeout
chain is measured so far and the new replicas evidence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The stand was recreated from a4ca9ef (make down, make run_docker) and
all evidence of the change was taken again on it: stand, distribution,
LB errors, test-persistence, test-failover for each replica (stop and
kill), stop-one refusals, make test and student_test through the LB,
test_isolated, failure_checks and bench self-checks.
New evidence closes scenarios verify found without proof: a real git
clone on non-default ports, a hung replica (docker pause: DOWN after
7.7 s, in-flight request 504), PostgreSQL outage behind the LB, replica
tests failing against one process, aborted responses via
driver.py kill-inflight. REPORT.md gets the corrected detection
formula, the measured cost of the 1 CPU limit for Argon2id and the new
rows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- stop-one.sh: an empty TARGET_INSTANCE_ID now exits 2 in any shell,
as the design states (${VAR:?} gave 1 under bash).
- driver.py kill-inflight: reproducible "aborted response is not
retried" scenario (logins in flight, replica killed via stop-one.sh).
- Detection time of a hung replica is fall x (inter + timeout check) =
8 s, not inter x fall + timeout check = 5 s (docker pause: 7.3 s);
fixed in haproxy.cfg, balancer.md and the design.
- The 1 CPU replica limit is kept on purpose; the compose comment and
the design now say that Argon2id uses more and logins are slower.
- load-balancing spec: API_REPLICAS is read by compose and Makefile.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Evidence from 4e0cc4e: stand ports, limits and per-service env,
X-Instance-Id distribution and scaling, LB-generated errors,
test-persistence through the LB, test-failover for each replica in stop
and kill modes, stop-one refusals, test runs and bench self-checks.
REPORT.md gains the LB architecture, timeout chain, limits, criteria
rows and defence answers. Coverage exclusions for restart now point to
the lab 3 persistence evidence; stale lab 2 pointers go to
KNOWN_ISSUES.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without init-state down HAProxy marked a slot UP as soon as Docker DNS
returned the container address, before uvicorn listened, so requests
after LB restart or scale-up hit a replica that was not ready.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
HAProxy 3.4 is the only published entry point; api runs as
API_REPLICAS replicas without ports, found through Docker DNS and
checked on /health/ready. Retries only on connection failure, LB errors
in the contract error format, resource limits and per-service secrets.
Adds scripts/stop-one.sh, LB-aware restart.sh, make replicas /
instances / lb_stats and replica tests driven by EXPECT_REPLICAS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
HAProxy as the single entry point, API replicas without published
ports, resource limits and per-service secrets, restart and stop-one
hooks, replica tests in student-tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- evidence for replay, conflict, replay through a new API process and
after restarting API and storages, keys table, key race,
X-Instance-Id in responses and logs, explicit and invalid
instance_id and key TTL, test runs
- stand evidence: make wipe_data leaves the lab 2 volume intact, build
from a fresh git clone passes smoke, bench restore stage sends with
Idempotency-Key
- REPORT.md: X-Instance-Id, idempotency design, IDEMPOTENCY_TTL_SECONDS
mapping, limitations and evidence rows
- tasks.md: mark the change tasks done, add follow-ups from verify
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- carry the lab 2 solution over and plan lab 3 (PLAN.md, OpenSpec
change instance-id-idempotency)
- rename the compose project to mtusi-lab3 and the image to :lab3
- add X-Instance-Id to every response and instance_id to every log
record ([app] instance_id, hostname by default)
- require Idempotency-Key on POST messages: keys table in Postgres,
key and message in one transaction, replay returns the same 201,
another body gets 409, TTL from [idempotency] ttl_seconds
- send the key from student-tests, k6 scenarios and the skill driver;
add idempotency, X-Instance-Id, key race and fault tests
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>