Spans, attributes, and outcomes
Trace structure, diagnostic fields, outcome rules, and the generated attribute inventory.
Spans, attributes, and outcomes
A span describes one operation, with attributes for its scope, outcome, and measured counts. Some fields explicitly identify the enclosing request or job; count units depend on the span kind.
Span structure
REST and hosted HTTP MCP use an HTTP entry span, while direct network gRPC calls use a gRPC server span. The internal REST/MCP-to-gRPC bridge preserves parentage without extra transport spans.
A retrieval trace can show a structure like this:
Request
Retrieval
Query preparation and embedding
Provider call
Attempts and retry waits
Search and result preparation
Postprocessing
Rerank or generation callThe request determines which operations appear. Parallel stages can overlap, so their duration sum can exceed the request duration. Provider observations also cover ingestion and provider checks outside retrieval. Use the generated span inventory for the current span names.
The inference proxy supplies transport traces only, without semantic generation spans or token counts. SQL spans require separate instrumentation.
REST and gRPC Decide calls add a gen_ai.decide child span with operation decide and
observation kind decision. It records the decision-model UUID, provider, requested/returned model,
duration and typed outcome. State, question names, rubrics, media and answers are excluded.
The Langfuse profile labels it as a generation; LangSmith labels it as an LLM run. These labels
do not change the typed decision API. Live decision ingestion and automatic pricing are unverified.
Request and resource filters
| Attribute | Type | Meaning |
|---|---|---|
goodmem.trace.name | Text | Enclosing GoodMem request or job name; not the remote caller's trace name |
goodmem.trace.owned | Boolean | GoodMem started the trace rather than joining the caller's trace |
goodmem.request.id | Text | Server request ID; separate from the trace ID |
goodmem.operation | Text | Operation on this span |
goodmem.request.operation | Text | Enclosing request operation on a child span |
goodmem.principal.id | UUID text | Authenticated principal |
goodmem.api_key.id | UUID text | Exact API-key identifier; never the key secret |
goodmem.resource.kind | Text | Resource kind for the operation |
goodmem.resource.id | UUID text | Validated resource identifier |
goodmem.retrieval.space_ids | Text array | Authorized search spaces; a bounded number of IDs |
goodmem.retrieval.accessible_space_count | Integer | Full authorized space count |
goodmem.retrieval.space_ids_truncated | Boolean | Whether the retained space list omits IDs |
goodmem.space.id | UUID text | Single authorized retrieval space, or the applicable resource/job space |
Retrieval space fields describe the authorized search scope. A truncated list can omit a space that the request searches.
If goodmem.retrieval.space_ids_truncated is true, a negative match cannot exclude an omitted space.
The generated inventory lists the vendor aliases for these fields.
Principal IDs remain resource metadata rather than backend application-user or session IDs.
Outcomes and errors
goodmem.outcome has four values:
| Value | Meaning |
|---|---|
success | Operation completed without a known failure |
failed | Operation failed, or no attempted logical item succeeded |
partial | Some work failed, or usable output required a fallback |
cancelled | Operation stopped before normal completion through cancellation |
| Failure field | Meaning |
|---|---|
goodmem.failure.category | What failed |
goodmem.failure.origin | Source of the failure, such as caller or provider |
goodmem.failure.expected | Positive identification of an expected rejection or intentional stop |
goodmem.status_code | GoodMem status where available |
goodmem.rpc.status_code | Canonical RPC result, such as DEADLINE_EXCEEDED |
rpc.grpc.status_code | Numeric gRPC transport status |
http.response.status_code | Numeric HTTP response status |
exception.type | Exception class where available; no raw message or stack trace |
Common categories include authentication, permission, timeout, and provider_failure. Other values can appear in later versions.
For cross-provider failure filters, use goodmem.failure.category; transport spans retain their native error.type conventions.
Result counts
A result count describes the output of the span that carries it, rather than one universal unit for the request. For example, retrieval counts describe emitted results, while vector-search counts describe candidate hits. Do not sum result counts across spans.
A stage reports a count only when the operation supplies one.
See the generated mappings for the fields that use the destination alias result_count.
Provider usage and retries
gen_ai.usage.input_tokens and gen_ai.usage.output_tokens contain the nonnegative token counts that the provider reports.
An absent counter means unknown usage; an explicit zero means reported zero.
Coverage follows the transport actually used by the configured provider:
| Embedding provider / route | Source of input tokens |
|---|---|
| OpenAI / vLLM | usage.prompt_tokens |
| Voyage | usage.total_tokens |
| Jina | usage.input_tokens when the field is present; otherwise usage.total_tokens |
| Cohere | meta.tokens.input_tokens, falling back to meta.billed_units.input_tokens |
| TEI dense and sparse | x-compute-tokens HTTP response header |
| DashScope native text | usage.total_tokens |
| DashScope OpenAI-compatible fallback | usage.prompt_tokens |
| DashScope multimodal endpoint (text inputs) | usage.input_tokens; explicit usage.output_tokens is recorded separately |
| Gemini Developer API (API key or ADC) / Google Cloud | usageMetadata.promptTokenCount |
This table is the authoritative source of provider field precedence. For Jina, a present but null
or invalid input_tokens field leaves input usage unknown; GoodMem does not substitute the total
in that case. Raw JSON is inspected before SDK numeric coercion.
These cover every provider supported by the embedding factory. The Gemini routes use the currently
supported Gemini embedding model's batchEmbedContents / embedContent operations. GoodMem reads
native metadata before the pinned SDK discards it or rounds Cloud counts through floating point.
It does not derive tokens from Google's billable character counts. Cohere's billed input tokens
are token units; search units and image counts are not. DashScope's multimodal total can include
output tokens, so it is not used as the input count.
See Cohere embedding usage,
Gemini embedding usage, and
DashScope multimodal usage.
| Reranker provider | Source of input tokens |
|---|---|
| Voyage | usage.total_tokens |
| Jina | usage.total_tokens |
| vLLM / llama.cpp | Shared Jina-compatible adapter: usage.total_tokens, when returned |
| DashScope (native and compatible) | usage.total_tokens |
| TEI | x-compute-tokens HTTP response header |
| Cohere | meta.tokens.input_tokens, if supplied; responses containing only meta.billed_units.search_units leave tokens unknown |
TEI's count covers the tokenized query/document pairs for the request. Local hosting does not imply zero usage or a known monetary price.
Cohere's meta.billed_units.search_units is exported separately as the numeric attribute
goodmem.rerank.cohere.search_units, with cohere_search_units metadata in both Langfuse and
LangSmith profiles. It preserves fractional quantities and explicit zero; absent, negative,
non-numeric, or non-finite values remain unknown. Search units are summed independently of token
counts on the logical rerank span. They are billing quantities, not token counts or currency, and
can support custom dashboards or cost calculations using the applicable provider rate.
See the provider contracts for vLLM,
TEI,
DashScope, and
Cohere.
The shared InferenceUsage accumulator publishes these counts; the generated inventory links one representative source per attribute.
These operations do not invent output-token counts or prices.
Each embedding/rerank response submits one immutable usage observation. Every observation accounts for input, output, total, model identity, and billing units; an omitted field is explicitly unknown. The shared raw-response inspector treats missing usage and inspection failures consistently and leaves SDK response consumption and transport configuration intact.
Counts are summed across decoded successful provider responses on the logical inference span, including batches. HTTP and retry child spans do not repeat token counts. Failed HTTP attempts do not supply counts; reported usage survives a later result-validation failure or cancellation when no batch work is abandoned. When concurrent execution abandons outstanding batches, provider usage and measurement exports for that invocation are suppressed: cancelled workers may still finish and report later. If any decoded response omits a valid count, that counter’s aggregate stays absent rather than reporting a partial total. Input and output completeness are independent. A call that makes no provider requests also leaves usage absent. See the generated inventory for usage fields, retry counters, and vendor aliases.
Jev, Laya and Clef decision calls publish valid usage.input_tokens and usage.output_tokens
on gen_ai.decide, with the same unknown-versus-zero rules. GoodMem does not derive a total.
Usage from a decoded successful HTTP response survives later answer-validation failure;
failed HTTP attempts and undecodable responses supply no counts.
Local providers and recognized destination attributes
For vLLM, Ollama, and llama.cpp chat streams, the provider request policy selected at client construction requests stream_options.include_usage=true when the
caller has not specified it. Explicit false and other stream options are preserved. This policy
is independent of tracing; the tracer observes usage without changing the request. A terminal
usage-only chunk supplies the totals; interrupted streams can still have unknown usage.
The OpenAI-compatible embedding transport also records gen_ai.response.model when every decoded
batch response supplies the same valid model. This preserves the actual served model alongside
the requested alias. The shared Jina-compatible rerank adapter likewise preserves a returned model,
including vLLM's model field. Both transports also record explicit usage.total_tokens as
gen_ai.usage.total_tokens, independently of the input/output split. Ollama currently has a chat provider in GoodMem, but no dedicated embedding
or rerank provider; an OpenAI-compatible embedding endpoint uses the OpenAI transport.
LLM spans emit the validated provider total as gen_ai.usage.total_tokens in every profile.
Client-observed streaming latency uses goodmem.llm.time_to_first_output_ms.
GoodMem does not emit or translate obsolete total-token or first-token aliases.
The compatibility profiles add these recognized attributes only when the source fact exists:
| Canonical fact | Langfuse export | LangSmith export |
|---|---|---|
goodmem.llm.cached_input_tokens | gen_ai.usage.cache_read.input_tokens | gen_ai.usage.cache_read_input_tokens, plus metadata |
goodmem.llm.reasoning_tokens | gen_ai.usage.reasoning.output_tokens | gen_ai.usage.details.reasoning_tokens |
gen_ai.response.model | Also langfuse.observation.model.name, selecting the returned model | Also ls_model_name in JSON metadata, with request model fallback |
gen_ai.provider.name | Canonical attribute retained | Also gen_ai.system, mapped to provider metadata |
Langfuse ingestion
treats cache reads and reasoning tokens as subsets of input/output, avoiding double counting.
LangSmith documents its reasoning, model, and provider mappings;
GoodMem uses the underscore cache-read alias verified against LangSmith ingestion, retaining the
canonical cached count in metadata as well. Cache reads remain a subset of total input tokens;
neither profile subtracts them from the canonical input count. LangSmith's pricing metadata explicitly selects the response model, falling
back to the request model; its invocation model alone is insufficient for price matching. Embedding
and rerank counts can appear in outputs.usage_metadata while top-level token totals remain zero;
those zeros do not mean zero consumption. Controlled backend pricing checks with synthetic
gpt-4o-mini usage verified discounted cache-read pricing in both destinations; this does not
establish other model catalogs, provider invoices, or embedding/rerank automatic pricing.
The neutral profile preserves canonical facts without destination aliases.
These counters depend on what the installed server returns. vLLM can return cached-token details when prompt-token details are enabled and reasoning-token details when a reasoning parser is configured. Ollama's OpenAI adapter can return prompt, completion, total, and cached prompt counts; GoodMem does not infer missing reasoning counts from output text.
TEI's usage source is x-compute-tokens for dense/sparse embeddings and reranking. Its response
measurements are captured separately below. GoodMem also accepts optional provider generation
measurements through the existing OpenAI-compatible SDK, without probing or upgrading runtimes:
| Provider extension | Canonical attributes | Availability |
|---|---|---|
llama.cpp root timings | goodmem.llama_cpp.timings. plus prompt_ms, predicted_ms, prompt_per_token_ms, predicted_per_token_ms, prompt_per_second, predicted_per_second | Completion and streaming SDK fixtures, including chunks without usage; live runtime/backend ingestion remains unverified |
Ollama root timings | goodmem.ollama.timings. plus prompt_ms, predicted_ms, prompt_per_token_ms, predicted_per_token_ms, prompt_per_second, predicted_per_second | Observed on Ollama 0.40.2 terminal streaming usage chunks; absent from tested non-streaming responses |
vLLM root metrics | goodmem.vllm.timings. plus time_to_first_token_ms, generation_time_ms, queue_time_ms, mean_itl_ms, tokens_per_second | Optional on completion and final streaming usage responses; requires server support/configuration |
All measurements are finite nonnegative float64 values; fractions and explicit zero survive, while
missing or malformed values remain absent. Repeated stream snapshots replace present fields without
summing; a missing later field does not erase a previously reported measurement. Early closure does
not consume a terminal chunk to obtain metrics. Explicit usage opt-out remains honored.
Both compatibility profiles expose these as diagnostic metadata named llama_cpp_, ollama_, or vllm_ plus the
field name. They have no claimed native latency or pricing mapping.
Provider timings do not replace span duration, client time to first visible output, or Langfuse's
completion-start timestamp. In vLLM v0.31.0,
server TTFT starts at scheduling, excluding queue time; generation time covers first-to-last output;
throughput includes the inference interval. Metrics require --enable-per-request-metrics, with
statistics logging where required; multiple choices suppress metrics in that version. The decoder
is SDK-fixture validated for this format, not E2E validated against a newer runtime. The audited
vLLM 0.11.0 runtime supplied no per-request metrics. Aggregate /metrics histograms are never
attributed to individual calls. Ollama native load_duration and total_duration are still unavailable
through the tested compatible responses; no native adapter or runtime upgrade is implied.
Compatibility with older Ollama builds that might reject stream_options remains unverified;
GoodMem does not negotiate runtime capabilities or send extra inference probes.
Explicit usage field coverage
The following matrix supplements the input precedence tables above. Counts are int64; every field
has independent completeness across successful responses. A missing, invalid, or overflowing count
suppresses only its aggregate. An explicit zero is preserved. These are provider-reported facts,
not sums inferred from other counters.
| Response field and extraction path | Canonical field | Backend interpretation |
|---|---|---|
Voyage embedding/rerank and native DashScope text/rerank usage.total_tokens | Both gen_ai.usage.input_tokens and gen_ai.usage.total_tokens for these input-only operations | Standard usage; destination-computed totals are not evidence of an emitted provider total |
Jina embedding raw usage.input_tokens (fallback to total only when absent), output_tokens, total_tokens | Independent input, output, total token fields | Raw inspection precedes SDK coercion; output and total remain valid even if input is malformed |
OpenAI-compatible embedding SDK usage prompt_tokens, optional extension completion_tokens, total_tokens | Independent input, output, total token fields | vLLM output zero is preserved; absent output remains unknown |
DashScope compatible text JSON usage.prompt_tokens, completion_tokens, total_tokens | Independent input, output, total token fields | Same presence rules as other compatible embeddings |
DashScope multimodal JSON usage.input_tokens, output_tokens, total_tokens | Independent input, output, total token fields | A total never supplies a missing multimodal input count |
TEI x-compute-tokens | Input only | No explicit total or output count is inferred |
TEI response measurements
TeiResponseMetadata decodes headers for dense, sparse and rerank. ProviderResponseMetrics
validates immutable observations; the invocation collector owns completeness and aggregation.
All numeric headers require one unambiguous, nonnegative decimal integer in signed 64-bit range.
Missing, duplicate, negative, fractional, and overflowing values are unknown; reported zero remains zero.
The logical span retains at most eight goodmem.provider.response events, in response completion
order, with receipt timestamps and no token counters or content. Each event carries
goodmem.provider.response.sequence and the valid raw measurements below. The summary accounts for
every decoded successful response, even when the bounded event buffer is full.
| Provider header | Event attribute under goodmem.provider.response. | Logical-span summary under the same prefix | Type / units |
|---|---|---|---|
x-compute-characters | compute_characters | compute_characters_total (complete sum) | int64 characters |
x-compute-time | compute_time_ms | compute_time_max_ms | int64 milliseconds |
x-total-time | total_time_ms | total_time_max_ms | int64 milliseconds |
x-tokenization-time | tokenization_time_ms | tokenization_time_max_ms | int64 milliseconds |
x-queue-time | queue_time_ms | queue_time_max_ms | int64 milliseconds |
x-inference-time | inference_time_ms | inference_time_max_ms | int64 milliseconds |
x-compute-type | compute_type | compute_type (only if every response agrees) | bounded provider label |
Each numeric summary is absent if any successful response lacks that valid field. Characters also
become unknown on sum overflow. count records successful responses; events_dropped counts samples
omitted by the eight-event buffer (SDK event limits can independently drop more). Failed HTTP attempts
supply neither counts nor measurements. Already observed responses survive later result-validation
failure or cancellation when no batch work is abandoned. Outstanding batches abandoned by concurrent
execution suppress these summaries and samples rather than exporting an incomplete snapshot.
These summaries do not claim completeness for unreported failed attempts.
TEI v1.9 reports component timings as per-item averages and total time as request elapsed time.
Maxima above are maxima of those reported values, not operation duration, summed work, or an average
across all batches. Components need not add to total. x-compute-type is a provider label; v1.9 can
report gpu+optimized even on CPU. See the versioned header source
and batch calculation.
Neutral export retains canonical events and summaries. Both compatibility profiles expose summaries
as metadata: provider_response_ plus the summary suffix, except compute_type uses
provider_compute_type. LangSmith receives numeric JSON metadata; Langfuse uses its existing string
metadata convention. These are diagnostic metadata, with no claimed native latency or cost mapping.
LangSmith additionally receives provider_response_samples, a structured metadata array containing
only the approved event measurements, sequence, and received_at_unix_ns as a decimal string to
preserve nanosecond precision. This retains at most eight samples and 2,048 UTF-8 bytes, as a prefix
in completion order. provider_response_samples_dropped, when present, counts export-projection
omissions separately from recorder provider_response_events_dropped and native SDK event drops.
This metadata is diagnostic detail, not a native LangSmith event timeline. Neutral and Langfuse
exports retain the original event representation without this projection.
They do not replace span duration or client-observed time to first output, which Langfuse receives
through langfuse.observation.completion_start_time.
Background jobs
Each execution attempt starts a new trace, separate from the original enqueue request.
Use goodmem.job.id to correlate attempts, or goodmem.job.attempt.id and goodmem.job.attempt.number to select one attempt.
Attempt duration covers execution only, without queue delay or time between durable retries.
Attribute limits
GoodMem bounds the number of attributes, text length, and array size per span.
The counters goodmem.telemetry.dropped_field_count and goodmem.attributes.dropped_count report omissions from these controls.
Generated span inventory
This table lists named spans from GoodMem source. HTTP route names and spans from dependency instrumentation vary with the request.
| Span | Source |
|---|---|
gen_ai.chat | InferenceTracing |
gen_ai.decide | InferenceTracing |
gen_ai.embeddings | InferenceTracing |
gen_ai.rerank | InferenceTracing |
goodmem.embedding.provider_attempt | EmbeddingAttemptTracing |
goodmem.embedding.retry_backoff | EmbeddingAttemptTracing |
goodmem.job.attempt | BgJobTracing |
goodmem.job.enqueue | MemoryServiceImpl |
goodmem.job.memory | BgJobTracing |
goodmem.openai.http_attempt | OpenAIHttpTracing |
goodmem.openai.retry_backoff | OpenAIHttpTracing |
goodmem.retrieval | RetrievalTracing |
goodmem.retrieval.fusion | RetrievalTracing |
goodmem.retrieval.hydration | RetrievalTracing |
goodmem.retrieval.plan | MemoryServiceImpl |
goodmem.retrieval.postprocess | MemoryServiceImpl |
goodmem.retrieval.query_embedding | RetrievalTracing |
goodmem.retrieval.vector_search | RetrievalTracing |
Generated attribute inventory
This inventory covers goodmem.* and gen_ai.* declarations in GoodMem source, without attributes from dependency instrumentation.
In goodmem.result.status_count.<code>, <code> is a lowercase gRPC status name.
Each span contains only the fields that its operation supplies.
The type column describes canonical OTLP values: int64 is an integer, and float64 is a floating-point number.
Langfuse scalar metadata aliases use strings. LangSmith metadata preserves canonical types.
The metadata columns list destination aliases; a dash marks fields without an alias, whose canonical attributes remain available. Standard GenAI fields also use native backend mappings.
Langfuse prefixes aliases with langfuse.observation.metadata., while LangSmith stores them in its JSON metadata attribute.
| Attribute | Type | Source | Langfuse metadata | LangSmith metadata |
|---|---|---|---|---|
gen_ai.operation.name | string | TelemetryAttributes | — | — |
gen_ai.provider.name | string | TelemetryAttributes | — | — |
gen_ai.request.choice.count | int64 | TelemetryAttributes | — | — |
gen_ai.request.frequency_penalty | float64 | TelemetryAttributes | — | — |
gen_ai.request.max_tokens | int64 | TelemetryAttributes | — | — |
gen_ai.request.model | string | TelemetryAttributes | — | ls_model_name |
gen_ai.request.presence_penalty | float64 | TelemetryAttributes | — | — |
gen_ai.request.seed | int64 | TelemetryAttributes | — | — |
gen_ai.request.temperature | float64 | TelemetryAttributes | — | — |
gen_ai.request.top_p | float64 | TelemetryAttributes | — | — |
gen_ai.response.finish_reasons | string[] | TelemetryAttributes | — | — |
gen_ai.response.model | string | TelemetryAttributes | — | ls_model_name |
gen_ai.usage.input_tokens | int64 | TelemetryAttributes | — | — |
gen_ai.usage.output_tokens | int64 | TelemetryAttributes | — | — |
gen_ai.usage.total_tokens | int64 | TelemetryAttributes | — | — |
goodmem.api_key.authority_mode | string | ApiObservation | — | — |
goodmem.api_key.ceiling_omitted | boolean | ApiObservation | — | — |
goodmem.api_key.ceiling_rule_count | int64 | ApiObservation | — | — |
goodmem.api_key.expiration_present | boolean | ApiObservation | — | — |
goodmem.api_key.id | string | ApiObservation | api_key_id | api_key_id |
goodmem.api_key.lifecycle | string | ApiObservation | — | — |
goodmem.api_key.status | string | ApiObservation | — | — |
goodmem.attributes.dropped_count | int64 | ApiObservation | — | — |
goodmem.audit_actor.id | string | TelemetryAttributes | — | — |
goodmem.authentication.mechanism | string | TelemetryAttributes | — | — |
goodmem.authentication.outcome | string | TelemetryAttributes | — | — |
goodmem.authorization.allowed_count | int64 | ApiObservation | — | — |
goodmem.authorization.audience | string | ApiObservation | — | — |
goodmem.authorization.not_allowed_count | int64 | ApiObservation | — | — |
goodmem.authorization.operation | string | ApiObservation | — | — |
goodmem.authorization.outcome | string | TelemetryAttributes | — | — |
goodmem.authorization.role | string | ApiObservation | — | — |
goodmem.authorization.selector | string | ApiObservation | — | — |
goodmem.authorization.target_id | string | ApiObservation | — | — |
goodmem.authorization.target_kind | string | ApiObservation | — | — |
goodmem.bootstrap.already_initialized | boolean | ApiObservation | — | — |
goodmem.credential.authority_mode | string | TelemetryAttributes | — | — |
goodmem.decision_model.id | string | TelemetryAttributes | decision_model_id | decision_model_id |
goodmem.drain.quiesced | boolean | ApiObservation | — | — |
goodmem.effective.interval_ms | int64 | ApiObservation | — | — |
goodmem.effective.limit | int64 | TelemetryAttributes | effective_limit | effective_limit |
goodmem.effective.max_in_flight | int64 | ApiObservation | — | — |
goodmem.effective.probe_count | int64 | ApiObservation | — | — |
goodmem.effective.timeout_ms | int64 | ApiObservation | — | — |
goodmem.effective.timeout_seconds | int64 | ApiObservation | — | — |
goodmem.embedder.id | string | TelemetryAttributes | embedder_id | embedder_id |
goodmem.embedding.adaptive_attempt_count | int64 | TelemetryAttributes | — | embedding_adaptive_attempt_count |
goodmem.embedding.attempt_number | int64 | TelemetryAttributes | — | — |
goodmem.embedding.batch_count | int64 | TelemetryAttributes | — | embedding_batch_count |
goodmem.embedding.configured_dimension | int64 | TelemetryAttributes | — | embedding_configured_dimension |
goodmem.embedding.context | string | TelemetryAttributes | embedding_context | embedding_context |
goodmem.embedding.dimensions | int64 | ApiObservation | — | — |
goodmem.embedding.distribution_type | string | ApiObservation | — | — |
goodmem.embedding.input_char_count | int64 | TelemetryAttributes | — | — |
goodmem.embedding.output_dimension | int64 | TelemetryAttributes | — | embedding_output_dimension |
goodmem.embedding.provider_attempt_count | int64 | TelemetryAttributes | — | embedding_provider_attempt_count |
goodmem.embedding.retry_backoff_count | int64 | TelemetryAttributes | — | embedding_retry_backoff_count |
goodmem.embedding.retry_count | int64 | TelemetryAttributes | — | embedding_retry_count |
goodmem.embedding.retry_elapsed_delay_ms | float64 | TelemetryAttributes | — | embedding_retry_elapsed_delay_ms |
goodmem.embedding.retry_requested_delay_ms | int64 | TelemetryAttributes | — | embedding_retry_requested_delay_ms |
goodmem.embedding.split_count | int64 | TelemetryAttributes | — | embedding_split_count |
goodmem.embedding.title_count | int64 | TelemetryAttributes | — | — |
goodmem.embedding.vector_type | string | TelemetryAttributes | — | — |
goodmem.enrollment.already_completed | boolean | ApiObservation | — | — |
goodmem.enrollment.bootstrap_eligible | boolean | ApiObservation | — | — |
goodmem.enrollment.status | string | ApiObservation | — | — |
goodmem.entry_surface | string | TelemetryAttributes | entry_surface | entry_surface |
goodmem.event_count | int64 | TelemetryAttributes | — | — |
goodmem.extension.status | string | ApiObservation | — | — |
goodmem.extension.type | string | ApiObservation | — | — |
goodmem.failure.category | string | ApiObservation | failure_category | failure_category |
goodmem.failure.expected | boolean | ApiObservation | failure_expected | failure_expected |
goodmem.failure.origin | string | ApiObservation | failure_origin | failure_origin |
goodmem.fusion.score_direction | string | TelemetryAttributes | — | fusion_score_direction |
goodmem.hnsw.defaults_used | boolean | TelemetryAttributes | — | hnsw_defaults_used |
goodmem.hnsw.ef_search | int64 | TelemetryAttributes | — | hnsw_ef_search |
goodmem.hnsw.iterative_scan | string | TelemetryAttributes | — | hnsw_iterative_scan |
goodmem.hnsw.max_scan_tuples | int64 | TelemetryAttributes | — | hnsw_max_scan_tuples |
goodmem.hnsw.scan_mem_multiplier | float64 | TelemetryAttributes | — | hnsw_scan_mem_multiplier |
goodmem.inference.completion_start_time | string | TracedLLMInference | — | — |
goodmem.inference.purpose | string | TelemetryAttributes | — | — |
goodmem.input_count | int64 | TelemetryAttributes | — | — |
goodmem.internal_rpc.cancelled_count | int64 | TelemetryAttributes | — | — |
goodmem.internal_rpc.count | int64 | TelemetryAttributes | — | — |
goodmem.internal_rpc.error_count | int64 | TelemetryAttributes | — | — |
goodmem.internal_rpc.partial_count | int64 | TelemetryAttributes | — | — |
goodmem.internal_rpc.success_count | int64 | TelemetryAttributes | — | — |
goodmem.job.attempt.id | int64 | TelemetryAttributes | job_attempt_id | job_attempt_id |
goodmem.job.attempt.number | int64 | TelemetryAttributes | job_attempt_number | job_attempt_number |
goodmem.job.completion.decision | string | BgJobTracing | — | — |
goodmem.job.completion.status_code | string | BgJobTracing | — | — |
goodmem.job.count | int64 | TelemetryAttributes | — | — |
goodmem.job.id | int64 | TelemetryAttributes | job_id | job_id |
goodmem.job.memory.result | string | BgJobTracing | — | — |
goodmem.job.memory.status_code | string | BgJobTracing | — | — |
goodmem.job.result | string | BgJobTracing | — | — |
goodmem.job.type | string | TelemetryAttributes | job_type | job_type |
goodmem.license.active_present | boolean | ApiObservation | — | — |
goodmem.license.reload_status | string | ApiObservation | — | — |
goodmem.lifecycle.state | string | ApiObservation | — | — |
goodmem.llama_cpp.timings.predicted_ms | float64 | TelemetryAttributes | llama_cpp_predicted_ms | llama_cpp_predicted_ms |
goodmem.llama_cpp.timings.predicted_per_second | float64 | TelemetryAttributes | llama_cpp_predicted_per_second | llama_cpp_predicted_per_second |
goodmem.llama_cpp.timings.predicted_per_token_ms | float64 | TelemetryAttributes | llama_cpp_predicted_per_token_ms | llama_cpp_predicted_per_token_ms |
goodmem.llama_cpp.timings.prompt_ms | float64 | TelemetryAttributes | llama_cpp_prompt_ms | llama_cpp_prompt_ms |
goodmem.llama_cpp.timings.prompt_per_second | float64 | TelemetryAttributes | llama_cpp_prompt_per_second | llama_cpp_prompt_per_second |
goodmem.llama_cpp.timings.prompt_per_token_ms | float64 | TelemetryAttributes | llama_cpp_prompt_per_token_ms | llama_cpp_prompt_per_token_ms |
goodmem.llm.budget_source | string | TelemetryAttributes | — | llm_budget_source |
goodmem.llm.cached_input_tokens | int64 | TelemetryAttributes | — | llm_cached_input_tokens |
goodmem.llm.choice_count | int64 | TelemetryAttributes | — | — |
goodmem.llm.chunk_count | int64 | TracedLLMInference | — | — |
goodmem.llm.configured_model | string | TelemetryAttributes | — | llm_configured_model |
goodmem.llm.context_limit | int64 | ApiObservation | — | — |
goodmem.llm.counts_truncated | boolean | TelemetryAttributes | — | — |
goodmem.llm.has_content | boolean | TelemetryAttributes | — | — |
goodmem.llm.has_refusal | boolean | TelemetryAttributes | — | — |
goodmem.llm.id | string | TelemetryAttributes | llm_id | llm_id |
goodmem.llm.reasoning_tokens | int64 | TelemetryAttributes | — | llm_reasoning_tokens |
goodmem.llm.sampling_params_applied | boolean | InferenceTracing | sampling_params_applied | sampling_params_applied |
goodmem.llm.system_message_mode | string | InferenceTracing | system_message_mode | system_message_mode |
goodmem.llm.time_to_first_output_ms | float64 | TelemetryAttributes | — | — |
goodmem.llm.tool_call_count | int64 | TelemetryAttributes | — | — |
goodmem.log_policy.active | boolean | ApiObservation | — | — |
goodmem.log_policy.active_from_present | boolean | ApiObservation | — | — |
goodmem.log_policy.active_until_present | boolean | ApiObservation | — | — |
goodmem.log_policy.clause_count | int64 | ApiObservation | — | — |
goodmem.log_policy.match_all | boolean | ApiObservation | — | — |
goodmem.mcp.is_error | boolean | TelemetryAttributes | — | — |
goodmem.mcp.method | string | TelemetryAttributes | — | — |
goodmem.mcp.tool | string | TelemetryAttributes | mcp_tool | mcp_tool |
goodmem.memory.id | string | TelemetryAttributes | memory_id | memory_id |
goodmem.memory.page_image_count | int64 | ApiObservation | — | — |
goodmem.memory.page_image_status | string | ApiObservation | — | — |
goodmem.memory.processing_status | string | ApiObservation | — | — |
goodmem.mutation.outcome | string | ApiObservation | — | — |
goodmem.observation.kind | string | MemoryServiceImpl | — | — |
goodmem.ocr.detected_format | string | ApiObservation | — | — |
goodmem.ocr.processing_ms | int64 | ApiObservation | — | — |
goodmem.ocr.queue_wait_ms | int64 | ApiObservation | — | — |
goodmem.ocr.render_ms | int64 | ApiObservation | — | — |
goodmem.ocr.wall_time_ms | int64 | ApiObservation | — | — |
goodmem.ollama.timings.predicted_ms | float64 | TelemetryAttributes | ollama_predicted_ms | ollama_predicted_ms |
goodmem.ollama.timings.predicted_per_second | float64 | TelemetryAttributes | ollama_predicted_per_second | ollama_predicted_per_second |
goodmem.ollama.timings.predicted_per_token_ms | float64 | TelemetryAttributes | ollama_predicted_per_token_ms | ollama_predicted_per_token_ms |
goodmem.ollama.timings.prompt_ms | float64 | TelemetryAttributes | ollama_prompt_ms | ollama_prompt_ms |
goodmem.ollama.timings.prompt_per_second | float64 | TelemetryAttributes | ollama_prompt_per_second | ollama_prompt_per_second |
goodmem.ollama.timings.prompt_per_token_ms | float64 | TelemetryAttributes | ollama_prompt_per_token_ms | ollama_prompt_per_token_ms |
goodmem.openai.body_completion | string | TelemetryAttributes | — | — |
goodmem.openai.connect_timeout_ms | int64 | TelemetryAttributes | — | — |
goodmem.openai.connection_attempt_count | int64 | TelemetryAttributes | — | — |
goodmem.openai.connection_duration_ms | int64 | TelemetryAttributes | — | — |
goodmem.openai.connection_failure_count | int64 | TelemetryAttributes | — | — |
goodmem.openai.failure_category | string | TelemetryAttributes | — | — |
goodmem.openai.http_attempt_count | int64 | TelemetryAttributes | — | openai_http_attempt_count |
goodmem.openai.http_exchange_count | int64 | TelemetryAttributes | — | — |
goodmem.openai.read_timeout_ms | int64 | TelemetryAttributes | — | — |
goodmem.openai.request_timeout_ms | int64 | TelemetryAttributes | — | — |
goodmem.openai.retry_backoff_count | int64 | TelemetryAttributes | — | openai_retry_backoff_count |
goodmem.openai.retry_elapsed_delay_ms | int64 | TelemetryAttributes | — | openai_retry_elapsed_delay_ms |
goodmem.openai.retry_requested_delay_ms | int64 | TelemetryAttributes | — | openai_retry_requested_delay_ms |
goodmem.openai.write_timeout_ms | int64 | TelemetryAttributes | — | — |
goodmem.operation | string | TelemetryAttributes | operation | operation |
goodmem.operation.cancelled | boolean | FailureTracing | — | — |
goodmem.operation.category | string | TelemetryAttributes | operation_category | operation_category |
goodmem.outcome | string | TelemetryAttributes | outcome | outcome |
goodmem.owner.id | string | TelemetryAttributes | — | — |
goodmem.ping.bytes_received | int64 | ApiObservation | — | — |
goodmem.ping.bytes_sent | int64 | ApiObservation | — | — |
goodmem.ping.error_count | int64 | ApiObservation | — | — |
goodmem.ping.http_status | int64 | ApiObservation | — | — |
goodmem.ping.last_probe_ok | boolean | ApiObservation | — | — |
goodmem.ping.request_count | int64 | ApiObservation | — | — |
goodmem.ping.rtt_ms | float64 | ApiObservation | — | — |
goodmem.ping.target_kind | string | ApiObservation | — | — |
goodmem.ping.timeout_count | int64 | ApiObservation | — | — |
goodmem.postprocess | boolean | TelemetryAttributes | — | — |
goodmem.postprocess.chronological_resort | boolean | TelemetryAttributes | — | chronological_resort |
goodmem.postprocess.fallback | boolean | TelemetryAttributes | — | postprocess_fallback |
goodmem.postprocess.filter_applied | boolean | TelemetryAttributes | — | postprocess_filter_applied |
goodmem.postprocess.kind | string | TelemetryAttributes | postprocess_kind | postprocess_kind |
goodmem.postprocess.max_results | int64 | TelemetryAttributes | — | postprocess_max_results |
goodmem.postprocess.record_projection | string | TelemetryAttributes | — | — |
goodmem.postprocess.relevance_threshold | float64 | TelemetryAttributes | — | relevance_threshold |
goodmem.postprocess.rerank_attempted | boolean | TelemetryAttributes | — | rerank_attempted |
goodmem.postprocess.rerank_succeeded | boolean | TelemetryAttributes | — | rerank_succeeded |
goodmem.postprocess.synthesis_attempted | boolean | TelemetryAttributes | — | synthesis_attempted |
goodmem.postprocess.synthesis_succeeded | boolean | TelemetryAttributes | — | synthesis_succeeded |
goodmem.principal.id | string | TelemetryAttributes | principal_id | principal_id |
goodmem.principal.type | string | TelemetryAttributes | — | — |
goodmem.provider.api_dialect | string | ApiObservation | — | — |
goodmem.provider.backend | string | ApiObservation | — | — |
goodmem.provider.modalities | string[] | ApiObservation | — | — |
goodmem.provider.response.compute_characters | int64 | TelemetryAttributes | — | — |
goodmem.provider.response.compute_characters_total | int64 | TelemetryAttributes | provider_response_compute_characters_total | provider_response_compute_characters_total |
goodmem.provider.response.compute_time_max_ms | int64 | TelemetryAttributes | provider_response_compute_time_max_ms | provider_response_compute_time_max_ms |
goodmem.provider.response.compute_time_ms | int64 | TelemetryAttributes | — | — |
goodmem.provider.response.compute_type | string | TelemetryAttributes | provider_compute_type | provider_compute_type |
goodmem.provider.response.count | int64 | TelemetryAttributes | provider_response_count | provider_response_count |
goodmem.provider.response.events_dropped | int64 | TelemetryAttributes | provider_response_events_dropped | provider_response_events_dropped |
goodmem.provider.response.inference_time_max_ms | int64 | TelemetryAttributes | provider_response_inference_time_max_ms | provider_response_inference_time_max_ms |
goodmem.provider.response.inference_time_ms | int64 | TelemetryAttributes | — | — |
goodmem.provider.response.queue_time_max_ms | int64 | TelemetryAttributes | provider_response_queue_time_max_ms | provider_response_queue_time_max_ms |
goodmem.provider.response.queue_time_ms | int64 | TelemetryAttributes | — | — |
goodmem.provider.response.sequence | int64 | TelemetryAttributes | — | — |
goodmem.provider.response.tokenization_time_max_ms | int64 | TelemetryAttributes | provider_response_tokenization_time_max_ms | provider_response_tokenization_time_max_ms |
goodmem.provider.response.tokenization_time_ms | int64 | TelemetryAttributes | — | — |
goodmem.provider.response.total_time_max_ms | int64 | TelemetryAttributes | provider_response_total_time_max_ms | provider_response_total_time_max_ms |
goodmem.provider.response.total_time_ms | int64 | TelemetryAttributes | — | — |
goodmem.provider.type | string | TelemetryAttributes | provider_type | provider_type |
goodmem.proxy.upstream.status_code | int64 | TelemetryAttributes | — | — |
goodmem.purge.attempt_count | int64 | ApiObservation | — | — |
goodmem.purge.dry_run | boolean | ApiObservation | — | — |
goodmem.purge.job_count | int64 | ApiObservation | — | — |
goodmem.purge.reference_count | int64 | ApiObservation | — | — |
goodmem.request.activation_present | boolean | ApiObservation | — | — |
goodmem.request.api_dialect | string | ApiObservation | — | — |
goodmem.request.authority_mode | string | ApiObservation | — | — |
goodmem.request.ceiling_rule_count | int64 | ApiObservation | — | — |
goodmem.request.chunk_length_measurement | string | ApiObservation | — | — |
goodmem.request.chunk_overlap | int64 | ApiObservation | — | — |
goodmem.request.chunk_size | int64 | ApiObservation | — | — |
goodmem.request.chunking_strategy | string | ApiObservation | — | — |
goodmem.request.client_provided_id | boolean | ApiObservation | — | — |
goodmem.request.content_bytes | int64 | ApiObservation | — | — |
goodmem.request.context_count | int64 | ApiObservation | — | — |
goodmem.request.context_limit | int64 | ApiObservation | — | — |
goodmem.request.continuation_present | boolean | ApiObservation | — | — |
goodmem.request.control_action | string | ApiObservation | — | — |
goodmem.request.credential_generation_mode | string | ApiObservation | — | — |
goodmem.request.cutoff_present | boolean | ApiObservation | — | — |
goodmem.request.dimensions | int64 | ApiObservation | — | — |
goodmem.request.distribution_type | string | ApiObservation | — | — |
goodmem.request.dpi | int64 | ApiObservation | — | — |
goodmem.request.dry_run | boolean | ApiObservation | — | — |
goodmem.request.embedder_count | int64 | ApiObservation | — | — |
goodmem.request.end_page | int64 | ApiObservation | — | — |
goodmem.request.expiration_present | boolean | ApiObservation | — | — |
goodmem.request.extension_type | string | ApiObservation | — | — |
goodmem.request.extract_page_images | boolean | ApiObservation | — | — |
goodmem.request.fetch_memory | boolean | ApiObservation | — | — |
goodmem.request.fetch_memory_content | boolean | ApiObservation | — | — |
goodmem.request.field_count | int64 | ApiObservation | — | — |
goodmem.request.fields | string[] | ApiObservation | — | — |
goodmem.request.filter_present | boolean | ApiObservation | — | — |
goodmem.request.format | string | ApiObservation | — | — |
goodmem.request.id | string | TelemetryAttributes | request_id | request_id |
goodmem.request.include_content | boolean | ApiObservation | — | — |
goodmem.request.include_credentials | boolean | ApiObservation | — | — |
goodmem.request.include_deleted | boolean | ApiObservation | — | — |
goodmem.request.include_enrollment_summary | boolean | ApiObservation | — | — |
goodmem.request.include_markdown | boolean | ApiObservation | — | — |
goodmem.request.include_processing_history | boolean | ApiObservation | — | — |
goodmem.request.include_raw_json | boolean | ApiObservation | — | — |
goodmem.request.include_revoked | boolean | ApiObservation | — | — |
goodmem.request.interval_ms | int64 | ApiObservation | — | — |
goodmem.request.item_count | int64 | ApiObservation | — | — |
goodmem.request.job_statuses | string[] | ApiObservation | — | — |
goodmem.request.key_lifecycle | string | ApiObservation | — | — |
goodmem.request.key_status | string | ApiObservation | — | — |
goodmem.request.key_view | string | ApiObservation | — | — |
goodmem.request.label_count | int64 | ApiObservation | — | — |
goodmem.request.label_selector_count | int64 | ApiObservation | — | — |
goodmem.request.label_update_strategy | string | ApiObservation | — | — |
goodmem.request.limit | int64 | TelemetryAttributes | requested_limit | requested_limit |
goodmem.request.max_in_flight | int64 | ApiObservation | — | — |
goodmem.request.media_type | string | ApiObservation | — | — |
goodmem.request.message_count | int64 | ApiObservation | — | — |
goodmem.request.min_chunk_size | int64 | ApiObservation | — | — |
goodmem.request.modalities | string[] | ApiObservation | — | — |
goodmem.request.name_filter_present | boolean | ApiObservation | — | — |
goodmem.request.new_owner_id | string | ApiObservation | — | — |
goodmem.request.operation | string | TelemetryAttributes | request_operation | request_operation |
goodmem.request.operation.category | string | TelemetryAttributes | request_operation_category | request_operation_category |
goodmem.request.owner_id | string | ApiObservation | — | — |
goodmem.request.page_index | int64 | ApiObservation | — | — |
goodmem.request.payload_size_bytes | int64 | ApiObservation | — | — |
goodmem.request.payload_type | string | ApiObservation | — | — |
goodmem.request.payload_variant | string | ApiObservation | — | — |
goodmem.request.postprocess | boolean | ApiObservation | — | — |
goodmem.request.probe_count | int64 | ApiObservation | — | — |
goodmem.request.processing_status | string | ApiObservation | — | — |
goodmem.request.provider_backend | string | ApiObservation | — | — |
goodmem.request.provider_type | string | ApiObservation | — | — |
goodmem.request.resource_id | string | ApiObservation | — | — |
goodmem.request.rotate_existing | boolean | ApiObservation | — | — |
goodmem.request.selector_kind | string | ApiObservation | — | — |
goodmem.request.sequence_limit | int64 | ApiObservation | — | — |
goodmem.request.space_count | int64 | ApiObservation | — | — |
goodmem.request.start_page | int64 | ApiObservation | — | — |
goodmem.request.subject_principal_id | string | ApiObservation | — | — |
goodmem.request.target_kind | string | ApiObservation | — | — |
goodmem.request.timeout_ms | int64 | ApiObservation | — | — |
goodmem.request.timeout_seconds | int64 | ApiObservation | — | — |
goodmem.request.wait_for_quiesce | boolean | ApiObservation | — | — |
goodmem.requested_size | int64 | TelemetryAttributes | — | requested_size |
goodmem.rerank.cohere.search_units | float64 | TelemetryAttributes | cohere_search_units | cohere_search_units |
goodmem.reranker.id | string | TelemetryAttributes | reranker_id | reranker_id |
goodmem.resource.deleted | boolean | ApiObservation | — | — |
goodmem.resource.id | string | TelemetryAttributes | resource_id | resource_id |
goodmem.resource.kind | string | TelemetryAttributes | resource_kind | resource_kind |
goodmem.resource.revoked | boolean | ApiObservation | — | — |
goodmem.response.message_count | int64 | ApiObservation | — | — |
goodmem.result.capability_count | int64 | ApiObservation | — | — |
goodmem.result.content_bytes | int64 | ApiObservation | — | — |
goodmem.result.count | int64 | TelemetryAttributes | result_count | result_count |
goodmem.result.deleted_count | int64 | ApiObservation | — | — |
goodmem.result.dpi | int64 | ApiObservation | — | — |
goodmem.result.embedder_count | int64 | ApiObservation | — | — |
goodmem.result.failure_count | int64 | TelemetryAttributes | failure_count | failure_count |
goodmem.result.informational_status_count | int64 | ApiObservation | — | — |
goodmem.result.media_type | string | ApiObservation | — | — |
goodmem.result.next_page_present | boolean | ApiObservation | — | — |
goodmem.result.page_index | int64 | ApiObservation | — | — |
goodmem.result.status_count.<code> | int64 | ApiObservation | — | — |
goodmem.result.success_count | int64 | TelemetryAttributes | success_count | success_count |
goodmem.result_count | int64 | TelemetryAttributes | result_count | result_count |
goodmem.retrieval.abstract_reply_count | int64 | TelemetryAttributes | — | — |
goodmem.retrieval.accessible_space_count | int64 | TelemetryAttributes | space_count | space_count |
goodmem.retrieval.boundary_count | int64 | TelemetryAttributes | — | — |
goodmem.retrieval.context_count | int64 | TelemetryAttributes | — | — |
goodmem.retrieval.effective_top_k | int64 | TelemetryAttributes | effective_top_k | effective_top_k |
goodmem.retrieval.fetch_content | boolean | TelemetryAttributes | — | — |
goodmem.retrieval.fetch_memory | boolean | TelemetryAttributes | — | — |
goodmem.retrieval.filter_count | int64 | TelemetryAttributes | — | — |
goodmem.retrieval.memory_definition_count | int64 | TelemetryAttributes | — | — |
goodmem.retrieval.required_embedder_count | int64 | TelemetryAttributes | — | — |
goodmem.retrieval.result_item_count | int64 | TelemetryAttributes | result_count | result_count |
goodmem.retrieval.space_ids | string[] | TelemetryAttributes | space_ids | space_ids |
goodmem.retrieval.space_ids_truncated | boolean | TelemetryAttributes | space_ids_truncated | space_ids_truncated |
goodmem.retrieval.status_codes | string[] | TelemetryAttributes | — | — |
goodmem.retrieval.status_count | int64 | TelemetryAttributes | — | — |
goodmem.rpc.status_code | string | ApiObservation | rpc_status_name | rpc_status_name |
goodmem.server.lifecycle_state | string | TelemetryAttributes | — | — |
goodmem.space.id | string | TelemetryAttributes | space_id | space_id |
goodmem.space_count | int64 | TelemetryAttributes | — | — |
goodmem.status_code | string | FailureTracing | — | — |
goodmem.streaming | boolean | TracedLLMInference | — | streaming |
goodmem.target.principal_id | string | ApiObservation | — | — |
goodmem.target.user_id | string | ApiObservation | — | — |
goodmem.telemetry.dropped_field_count | int64 | TelemetryAttributes | — | — |
goodmem.trace.name | string | TraceAttribution | — | — |
goodmem.trace.owned | boolean | TelemetryAttributes | — | — |
goodmem.vector.dimension | int64 | TelemetryAttributes | — | vector_dimension |
goodmem.vector.metric | string | TelemetryAttributes | — | vector_metric |
goodmem.vllm.timings.generation_time_ms | float64 | TelemetryAttributes | vllm_generation_time_ms | vllm_generation_time_ms |
goodmem.vllm.timings.mean_itl_ms | float64 | TelemetryAttributes | vllm_mean_itl_ms | vllm_mean_itl_ms |
goodmem.vllm.timings.queue_time_ms | float64 | TelemetryAttributes | vllm_queue_time_ms | vllm_queue_time_ms |
goodmem.vllm.timings.time_to_first_token_ms | float64 | TelemetryAttributes | vllm_time_to_first_token_ms | vllm_time_to_first_token_ms |
goodmem.vllm.timings.tokens_per_second | float64 | TelemetryAttributes | vllm_tokens_per_second | vllm_tokens_per_second |