Каждый tool-output хранился ДВАЖДЫ: в metadata.parts (assistantParts) И в
tool_calls (serializeSteps). При 50-шаговом ране с outputs по 50–200 KB это
127–510 МБ записи в Postgres за ход (+WAL/TOAST/dead tuples), т.к. onStepFinish
переписывает всю строку. Копия в parts — та, что реально реплеится модели и
рендерится UI/markdown-экспортом, так что копия в трейсе была чистым дублем.
Новый формат элементов tool_calls (v2), парно на каждый вызов:
{toolName, input} — вызов
{toolName, ok: true} — успех (БЕЗ output)
{toolName, error, kind: 'thrown'} — брошенный tool-error
{toolName, error, kind: 'interrupted'} — прерван mid-step (abort/restart)
kind обязателен: синтетический «Tool call did not complete.» при прерывании иначе
неотличим от реального hard-fail и загрязняет error-rate. Различие структурное
(errorsById-хит против синтетической ветки), НЕ per-tool классификатор — soft-
маркеры в трейс не выносятся (остаются в metadata.parts).
metadata.toolTraceVersion: 2 — маркер эры; старые строки НЕ мигрируются
(перезапись гигантских jsonb — тот самый WAL-чарн). serializeSteps пейрит
результаты/ошибки по toolCallId (как assistantParts); общая константа
TOOL_CALL_INCOMPLETE_TEXT держит текст реплея и трейса в синхроне.
docs/reading-ai-logs.md переписан dual-shape: ветвление по toolTraceVersion,
soft-анализ v2 через metadata.parts, правило «не сравнивать агрегаты через границу
эр». UI action-log и markdown-экспорт читают только parts — не затронуты.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
21 KiB
Reading the AI dialog logs (how agents call tools, and where they fail)
How to inspect the agent conversation history in the database — and, more
importantly, the one non-obvious trap that will make you report the wrong
answer: the persisted history silently hides hard tool failures. Written from
real pain (a "which tools fail most?" analysis that confidently answered
"patchNode: 0 errors" while the UI was visibly full of red patchNode failures).
Read the Gotchas section before you trust any error count.
TWO ERAS — check the marker first. The
tool_callsshape changed in #490 (trace v2). A row written by v2 carriesmetadata.toolTraceVersion = 2; older rows have no such key. The two shapes store DIFFERENT things (v2 dropped the tool OUTPUT from the trace), so every query below is dual-shape — branch on the marker. Never compare an aggregate or trend across the era boundary: a metric jump on the cut-over week is an artifact of the shape change, not a behavior change.
TL;DR
- Agent chats live in Postgres, DB
docmost, tablesai_chat_*. - Era marker:
metadata.toolTraceVersion = 2⇒ v2 (#490) row; absent ⇒ legacy row. - Each tool invocation is stored as two consecutive array elements — a
tool-callpart then an OUTCOME part — so naive counting double-counts.- v2 (#490): outcome is
{toolName, ok: true}on success, or{toolName, error, kind: 'thrown'|'interrupted'}on failure. The tool OUTPUT is NOT intool_callsany more — it lives once inmetadata.parts(this removed a hundreds-of-MB-per-run write duplication). Soft-failure analysis therefore readsmetadata.parts, nottool_calls. - legacy: outcome is
{toolName, output}on success; a thrown failure is a{toolName, error}element only on rows after #407, and is dropped entirely (silent orphan) on pre-#407 rows.
- v2 (#490): outcome is
- A tool that throws writes no result part. In v2 it is a
{error, kind:'thrown'}element; an interrupted/aborted call is a distinct{error, kind:'interrupted'}.isError/success=falsescans read the output and so under-report thrown failures in every era. - To find where agents fail: (1) soft-failure markers in
metadata.partsoutputs (v2) /tool_callsoutputs (legacy), (2) theerror/kindfields for thrown failures (v2 + post-#407), (3) server logs / the live UI for full stack traces.
Where the data lives
Host island.lc (10.31.40.120), container gitmost-postgresql
(pgvector/pgvector:pg18), database docmost.
ssh island.lc
# one-off query:
docker exec gitmost-postgresql psql -U docmost -d docmost -P pager=off -c "SELECT ..."
# interactive:
docker exec -it gitmost-postgresql psql -U docmost -d docmost
The main app container is gitmost (DATABASE_URL=postgresql://docmost:...@db:5432/docmost).
All workspaces (vvzvlad / wb / asakusa / …) share this single database — they
are rows in workspaces, not separate deployments.
Relevant tables
| Table | What it holds |
|---|---|
ai_chats |
one row per conversation (title, role_id, page_id, creator_id) |
ai_chat_messages |
every message; tool calls live in tool_calls jsonb |
ai_chat_runs |
one row per agent run (turn): status, error, step_count |
ai_agent_roles |
agent definitions (instructions, model_config) |
ai_mcp_servers |
configured MCP tool servers per workspace |
ai_chat_messages columns that matter: role (user | assistant — there is no
separate tool role), content (text), tool_calls (jsonb array), metadata
(jsonb, holds run error + rendered parts), status, tsv (full-text index).
Era marker — check this before every query
-- how many rows are in each era?
SELECT COALESCE((metadata->>'toolTraceVersion'), 'legacy') AS era, count(*)
FROM ai_chat_messages
WHERE role = 'assistant' AND jsonb_typeof(tool_calls) = 'array'
GROUP BY 1 ORDER BY 2 DESC;
toolTraceVersion = '2'→ v2 (#490): outcome flags, no output in the trace.NULL('legacy') → pre-#490: outcome carries the tooloutputinline.
Do not trend a metric across the cut-over. The shape change alone shifts counts
(e.g. "elements with output" collapses to zero for v2), so a week that straddles
the boundary shows an artifact, not a behavior change. Segment by era, or restrict to
one era, before comparing.
How tool calls are stored — READ THIS
Tool calls are not one-object-per-call. Each logical invocation is split into
two consecutive elements of the tool_calls array — a call then an outcome.
The outcome shape is era-dependent:
# v2 (#490) — metadata.toolTraceVersion = 2
index 0: { "toolName":"getPage", "input":{...} } ← call (has input)
index 1: { "toolName":"getPage", "ok":true } ← success (NO output here)
or : { "toolName":"getPage", "error":"…", "kind":"thrown" } ← threw
or : { "toolName":"getPage", "error":"…", "kind":"interrupted" } ← aborted mid-step
# legacy — no toolTraceVersion
index 0: { "toolName":"getPage", "input":{...} } ← call (has input, NO output)
index 1: { "toolName":"getPage", "output":{...} } ← success (has output)
or : { "toolName":"getPage", "error":"…" } ← threw (post-#407 only)
The keys that can appear: toolName, input (call), and on the outcome — v2:
ok or error+kind; legacy: output or (post-#407) error. There is
no state, no errorText, no type in tool_calls (those live on metadata.parts).
Consequences:
- Real invocation count — count the OUTCOME elements, not every element (else you
double-count): v2 = elements with
okorerror; legacy = elements withoutputorerror. - Pairing: a call (
input) is followed by its outcome.toolNameis on both, so you can group by tool on either. In v2 thekindfield separates a real hard-fail (thrown) from an aborted call (interrupted) — a distinction legacy rows cannot make (both are orphans; see below). - The tool OUTPUT is only in
metadata.partson v2 rows. To inspect what a tool returned (soft-error markers, page bodies) on a v2 row, read the parts (part->>'type' LIKE 'tool-%',part->>'state' = 'output-available',part->'output'), nottool_calls.
The two classes of failure (and which the DB can see)
1. Soft failures — tool RAN and returned an error-shaped result → PERSISTED ✅
These are visible in the tool output — on v2 rows in metadata.parts (the
output-available part's output), on legacy rows in the tool_calls outcome
element's output. The marker differs per tool:
| Tool(s) | Error marker in output |
|---|---|
editPageText |
failed is a non-empty array of {find, reason} (e.g. text not found in the document, matches N times — provide a longer fragment or set replaceAll). Also a soft warning when the find string contained markdown that only matched after stripping. |
semanticSearch |
{ "unavailable": true, "reason": "semantic search unavailable" } (feature/infra, not the agent's fault) |
MCP passthrough (Habr_*, some Search_*) |
output is an array (raw MCP content) whose text starts with Error executing tool … validation error … |
| generic | output.isError = true or output.success = false |
Note editPageText returns failed: [] on success — filtering on the presence
of the key gives false positives; filter on non-empty.
2. Hard failures — tool THREW → PERSISTED ✅
When a tool throws (the classic one is patchNode / insertNode / tableUpdateCell
→ Failed to encode document to Yjs (fromJSON): Unknown node type: undefined), the
runtime writes no tool-result part — the failure is an ai@6 tool-error content
part. How that lands in tool_calls depends on the era:
- v2 (#490): a
{toolName, error, kind:'thrown'}outcome element. An interrupted / aborted mid-step call is a distinct{toolName, error:'Tool call did not complete.', kind:'interrupted'}element — so you can tell a real hard-fail from an abort directly, without the orphan heuristic. Querykind = 'thrown'. - post-#407 legacy: a
{toolName, error}element (nokind) right after the call. - pre-#407 legacy: the error is dropped — a silent orphan (a
callwith nooutputand noerror).
The same real error text is replayed to the model on the next turn (an output-error
part with the real errorText, from metadata.parts), in every era.
Cutover caveat. Only pre-#407 legacy rows need the orphan proxy: an orphan is a
tool-call with no matching outcome. Orphans there also appear when a run is aborted
mid-flight (server restart), so a high-volume tool (createComment, searchInPage,
Search_web_search) shows orphans from aborts, not real errors. Treat the orphan gap as
an upper bound and cross-check the tool: a gap on a structural editor (patchNode,
insertNode, updatePageJson, transformPage) is almost certainly a thrown Yjs-encode
error; a gap on createComment is mostly aborts. On v2 rows this ambiguity is gone
— kind labels each outcome.
3. Run-level failures → ai_chat_runs
status ∈ succeeded | aborted | failed | running; error holds the text. Seen in
the wild: Run interrupted by a server restart. (aborts) and
Failed after N attempts. Last error: The service may be temporarily overloaded
(LLM provider 529). These are infra/provider, not agent tool misuse.
Ready-to-use queries
Run all of these via docker exec gitmost-postgresql psql -U docmost -d docmost -P pager=off -c "…".
Real invocation count per tool (outcome parts only — the correct denominator).
Dual-shape: a v2 outcome has ok or error; a legacy outcome has output or error:
SELECT elem->>'toolName' AS tool, count(*) AS calls
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE jsonb_typeof(m.tool_calls) = 'array'
AND (elem ? 'ok' OR elem ? 'output' OR elem ? 'error')
GROUP BY 1 ORDER BY 2 DESC;
Soft errors per tool. The soft-error marker lives in the tool OUTPUT — which on
v2 rows is in metadata.parts, on legacy rows is in the tool_calls outcome
element. This query UNIONs both eras, projecting each output as o:
WITH res AS (
-- v2 (#490): output is in metadata.parts (output-available tool parts)
SELECT part->>'type' AS tool, part->'output' AS o
FROM ai_chat_messages m, jsonb_array_elements(m.metadata->'parts') part
WHERE (m.metadata->>'toolTraceVersion') = '2'
AND part->>'type' LIKE 'tool-%' AND part->>'state' = 'output-available'
UNION ALL
-- legacy: output is inline in the tool_calls outcome element
SELECT elem->>'toolName' AS tool, elem->'output' AS o
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE (m.metadata->>'toolTraceVersion') IS NULL
AND jsonb_typeof(m.tool_calls) = 'array' AND elem ? 'output'
)
SELECT tool, count(*) AS calls,
sum(COALESCE(
(o->>'isError') = 'true'
OR (o->>'success') = 'false'
OR (jsonb_typeof(o->'failed') = 'array' AND o->'failed' <> '[]'::jsonb)
OR (o->>'unavailable') = 'true'
OR o::text ~* 'error executing tool|validation error'
, false)::int) AS soft_errors
FROM res GROUP BY tool HAVING sum(COALESCE(
(o->>'isError') = 'true' OR (o->>'success') = 'false'
OR (jsonb_typeof(o->'failed') = 'array' AND o->'failed' <> '[]'::jsonb)
OR (o->>'unavailable') = 'true' OR o::text ~* 'error executing tool|validation error'
, false)::int) > 0
ORDER BY soft_errors DESC;
Note the v2 tool label is the part type (tool-editPageText); strip the tool-
prefix if you join it against the legacy toolName.
editPageText failure reasons (the most common real agent mistake — bad find).
Same dual-shape output source:
WITH res AS (
SELECT part->'output' AS o
FROM ai_chat_messages m, jsonb_array_elements(m.metadata->'parts') part
WHERE (m.metadata->>'toolTraceVersion') = '2'
AND part->>'type' = 'tool-editPageText' AND part->>'state' = 'output-available'
UNION ALL
SELECT elem->'output' AS o
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE (m.metadata->>'toolTraceVersion') IS NULL
AND jsonb_typeof(m.tool_calls) = 'array'
AND elem->>'toolName' = 'editPageText' AND elem ? 'output'
)
SELECT f->>'reason' AS reason, count(*)
FROM res, jsonb_array_elements(o->'failed') f
WHERE jsonb_typeof(o->'failed') = 'array'
GROUP BY 1 ORDER BY 2 DESC;
Hard errors — persisted error field per tool (v2 + post-#407 rows) — thrown tool
failures carry their real reason, so query them directly. On v2 rows exclude the
interrupted kind so an aborted call is not counted as a hard-fail:
SELECT elem->>'toolName' AS tool, count(*) AS thrown_errors,
min(elem->>'error') AS sample_error
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE jsonb_typeof(m.tool_calls) = 'array' AND elem ? 'error'
-- v2 rows label the kind; a legacy error element has no kind (count it).
AND COALESCE(elem->>'kind', 'thrown') = 'thrown'
GROUP BY 1 ORDER BY 2 DESC;
Aborted mid-step calls on v2 rows are a distinct, directly countable population:
SELECT elem->>'toolName' AS tool, count(*) AS interrupted
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE jsonb_typeof(m.tool_calls) = 'array' AND elem->>'kind' = 'interrupted'
GROUP BY 1 ORDER BY 2 DESC;
Hard-error proxy for OLD rows (pre-#407) — orphan gap per tool, WITH a spread column
(call parts minus outcome parts, plus how many distinct chats the gap is spread across).
This is needed ONLY for pre-#407 legacy rows (v2 and post-#407 rows carry the error /
kind directly — use the queries above). The WHERE restricts to the legacy era so v2
rows (where an ok outcome is not an output) never produce phantom orphans:
WITH parts AS (
SELECT m.chat_id, elem->>'toolName' AS tool,
(elem ? 'input' AND NOT (elem ? 'output') AND NOT (elem ? 'ok')) AS is_call,
(elem ? 'output' OR elem ? 'error' OR elem ? 'ok') AS is_result
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE jsonb_typeof(m.tool_calls) = 'array' AND m.role = 'assistant'
AND (m.metadata->>'toolTraceVersion') IS NULL
),
per_chat AS (
SELECT tool, chat_id, sum(is_call::int) - sum(is_result::int) AS gap
FROM parts GROUP BY tool, chat_id
)
SELECT tool,
sum(gap) FILTER (WHERE gap > 0) AS missing_results,
count(*) FILTER (WHERE gap > 0) AS chats_spread, -- disambiguates!
max(gap) AS worst_single_chat
FROM per_chat GROUP BY tool
HAVING sum(gap) FILTER (WHERE gap > 0) > 0
ORDER BY missing_results DESC;
The is_result predicate counts an error element as a paired result too, so on new
rows a persisted thrown error no longer inflates the orphan gap; a remaining gap is an
aborted/interrupted call.
On OLD rows, missing_results mixes thrown errors AND aborted/interrupted runs — you
cannot split them from output alone (a positional "what follows the orphan" heuristic
breaks on parallel tool batches, which persist as call,call,…,result,result). Use
chats_spread to disambiguate:
- spread across many chats (e.g.
createComment96 over 29 chats) → a systemic real error (here: inline-comment anchor text not found on the page). - concentrated in one chat (e.g.
searchInPage55, of which 51 in a single chat) → one runaway/aborted session, not a real per-call error — discount it. - a gap on a structural editor (
patchNode,insertNode,tableUpdateCell,updatePageJson,transformPage) is almost always a thrown Yjs-encode error.
Run-level failures:
SELECT status, count(*), min(error) AS sample_error
FROM ai_chat_runs GROUP BY status ORDER BY 2 DESC;
Full-text search across messages. The tsv GIN index is built as
to_tsvector('english', unaccent(content)) — so it stems English but not
Russian (Russian lexemes are stored unstemmed, so only exact word forms match).
Most content here is Russian, so prefer ILIKE for substring search:
-- Russian / substring — reliable:
SELECT chat_id, left(content, 120)
FROM ai_chat_messages WHERE content ILIKE '%иранск%' LIMIT 20;
-- English phrase — can use the index:
SELECT chat_id, left(content, 120)
FROM ai_chat_messages
WHERE tsv @@ websearch_to_tsquery('english', 'some phrase') LIMIT 20;
Don't blow up your context
Tool outputs embed full page content and search payloads (hundreds of KB per row).
On legacy rows they are in tool_calls; on v2 rows they moved to
metadata->'parts' (the tool_calls trace itself is now small). Never SELECT tool_calls / metadata (or jsonb_pretty(...)) raw — project just the keys you need
and truncate:
-- v2: outputs live in metadata.parts
SELECT part->>'type',
left(regexp_replace((part->'output')::text, '\s+', ' ', 'g'), 200)
FROM ai_chat_messages m, jsonb_array_elements(m.metadata->'parts') part
WHERE (m.metadata->>'toolTraceVersion') = '2'
AND part->>'state' = 'output-available' LIMIT 5;
-- legacy: outputs live in tool_calls
SELECT elem->>'toolName',
left(regexp_replace((elem->'output')::text, '\s+', ' ', 'g'), 200)
FROM ai_chat_messages m, jsonb_array_elements(m.tool_calls) elem
WHERE elem ? 'output' LIMIT 5;
Server logs & live UI (for the error text the DB drops)
docker logs -f --tail=100 gitmost # main app
docker compose -p gitmost logs -f --tail=100 # whole stack
Logging is json-file, max-size=10m max-file=5 → ~50 MB retained, then rotated,
and wiped on container recreate. Thrown-tool error text is persisted — in the
error field of tool_calls (v2 kind:'thrown' / post-#407 legacy) — so you no longer
depend on live logs for it. Logs/live UI remain useful for pre-#407 rows (whose
thrown errors were dropped) and for full stack traces beyond the truncated stored
message. A per-tool tool_calls_total{tool,status} metric to VictoriaMetrics is still a
possible future add for aggregate dashboards.
Gotchas checklist
- Check
metadata.toolTraceVersionfirst. v2 (= 2) has no output intool_calls; legacy has it inline. Never trend a metric across the era boundary. - Counting every
tool_callselement → overcount. Count OUTCOME elements — v2:okorerror; legacy:outputorerror— never both call+outcome as invocations. isError/success=false≈ 0 does not mean "no errors" — thrown errors are anerrorelement (v2kind:'thrown'/ post-#407), not in the output.- v2: soft-error markers (the tool output) are in
metadata.parts, NOTtool_calls. Legacy: they are in thetool_callsoutcomeoutput. - v2:
kindsplits a real hard-fail (thrown) from an aborted call (interrupted) directly — no orphan heuristic needed. The orphan gap is a pre-#407-legacy-only proxy. editPageText.failedis[]on success — test for non-empty, not presence.abortedruns = server restarts,failedruns = provider overload — not agent mistakes.- Never dump a raw
tool_callsormetadata.partscell — outputs are hundreds of KB. - Logs are ephemeral (≤50 MB, wiped on recreate) — grab pre-#407 hard-error text live.
Snapshot (2026-07-07, illustrative — rerun the queries for current numbers)
All rows in this snapshot predate #490, so they are legacy-era (outputs inline in
tool_calls, orphan proxy for thrown errors). Do not trend these numbers against v2 rows — segment bytoolTraceVersionfirst.
- 226 chats, 732 messages, 46 runs; ~4 400 real tool invocations.
- Soft errors (persisted):
editPageText4/79 (bad/non-uniquefind) + 9 markdown-in-findwarnings;semanticSearch3/4 (unavailable);Habr_update_draft_from_docmost1/2 (docsent as object, not string). - Missing-result proxy, read WITH the spread column:
- Systemic (spread) → real errors:
createComment96 over 29 chats (comment anchor text not found — the biggest real error hotspot);editPageText31 over 12 chats (+ the 4 soft above); structural-editor Yjs throwsinsertNode10 /updatePageContent9 /tableUpdateCell6 /patchNode5 /updatePageJson2 /transformPage2. - Concentrated → NOT real errors:
searchInPage55 (51 in one chat);Search_web_search15 &Search_searxng_web_search6 (timeouts/aborts in long research sessions).
- Systemic (spread) → real errors:
- Runs: 34 succeeded, 10 aborted (server restart), 1 failed (provider overload).