Query a single trajectory - Simple Log Service - Alibaba Cloud Documentation Center

Prerequisites

Data in the selected format is ingested, and the corresponding trajectory processing results and query indexes are available. If raw data has just appeared, allow time for processing.

Find a trajectory

Open the destination SessionStore and select Single Trajectory.
Set the time range.
Enter an SLS search condition in the search box and click Search & Analyze. If no condition is required, use *.
Alternatively, open the filter panel, set the conditions, and then click Search.
In the Trajectory List, click the ID or Details of the destination trajectory.

By default, the list displays 20 entries per page. You can select 20, 50, or 100 entries per page. A new query with a relative time range is updated to the new time window, whereas pagination uses the time window of the submitted query.

Common query conditions

Purpose	Example
Find a trajectory by outer trajectory ID	`trajectory_id:"demo-trajectory-001"`
Find a trajectory by trace ID	`trace_id:"demo-trace-001"`
Find trajectories that belong to the same session	`session_id:"demo-session-001"`
Find trajectories that contain a specific agent	`agent_names:"main-agent"`
Find trajectories that contain a specific tool	`tool_names:"search_order"`

You can use AND to combine multiple conditions. Example:

session_id:"demo-session-001" AND tool_names:"search_order"

Note: The search box of the trajectory list is used for search conditions, not for entering complete SQL analytic statements. To query log records or use the SQL or SPL statements that are supported by the current service, switch to Log View. Pasting an ID alone is not equivalent to an exact query by the ID field.

The trajectory list and Log View retain their own query and time states, and the same conditions are not necessarily retained after you switch between them. To compare the same query, enter the same search conditions after you switch, check that the absolute time range is the same, and then click Search & Analyze.

Use structured filters

The filter panel provides conditions such as agent, model, tool, number of steps, duration, trajectory or session identifier, and input and output content. You can set the duration in seconds or minutes as a greater-than, less-than, or range condition.

Multiple model or tool values in the same dimension are combined by using OR, and different dimensions are combined by using AND. The Representative Agent filter actually finds trajectories that contain the selected agent, and is not limited to the single representative agent that is displayed on the card.

The candidate list is obtained from samples within the current time range and is not a complete list. For values that do not appear in the candidate list, you can enter the actual identifiers to query.

Important

Clicking Search in the filter panel generates and submits new query conditions, instead of automatically appending the conditions to the end of your manually written query. After you manually modify a query, check the conditions that you want to retain. Resetting the filters does not mean that the query is run again.

Semantic matching of inputs and outputs

If the destination fields of the current resource have embeddings and vector indexes, you can select semantic matching in the input or output conditions. If no corresponding index exists, use keyword matching instead.

The semantic distance ranges from 0 to 1. A smaller distance indicates higher similarity. A smaller threshold makes the condition stricter instead of increasing recall. Whether semantic matching is available and whether the destination data can be retrieved also depend on the indexes and the data processing status.

The current system trajectory index is read-only on the SessionStore page, and you cannot enable its vector index on this page. If required, contact the service administrator to confirm the capability.

Read trajectory details

The details page consists of three linked sections.

Section	Operation	Content
Trajectory topology	Select an agent node	The agents and their relationships that are explicitly provided in the current snapshot.
Execution steps	Select a step or use step search	The execution events of the selected node. Step search queries only the content that is currently loaded.
Step details	Switch between Overview, Request Parameters, Result, and Raw Data	The inputs and outputs, tool parameters, results, and raw structure of the current step.

You can adjust the width of a section or maximize a section to read long inputs and outputs.

Models, tools, and request parameters

The Overview tab displays content such as models, tokens, duration, tool calls, and body text based on the collected data. The Request Parameters tab is currently used to view the collected tool parameters, and does not necessarily contain all parameters of model requests.

If Request Parameters Not Collected is displayed or a field is empty, it only indicates that the current data does not provide the information. Do not interpret missing values as no parameters, no consumption, or no calls in the business.

Historical context

Historical context indicates messages that are carried over from previous trajectories, and is usually displayed in collapsed groups. It is not evidence that the current trajectory was executed again, and cannot be simply added to the actual execution steps as the number of calls in this execution.

Agent relationships and time range

Aggregation containers in the topology are used to organize data and do not represent actual agent calls. The system preferentially displays relationships that have explicit evidence. If the relationship data is incomplete, a message may appear or the display may be downgraded.

The timeline displays the time range and step coverage that are visible in the current snapshot. The intervals and coverage between steps cannot be directly used as the model computation time or the complete agent lifecycle. Downgraded indicates that the display quality of the data is limited, and does not mean that the task execution failed.

View the trajectory snapshot JSON

Use Trajectory Snapshot JSON to view the entire record that is currently selected. This is suitable for checking fields or submitting anonymized diagnostic information. This entry is not a list of historical versions and cannot be used to modify trajectory content.

Understand the differences between the list and details

The same trajectory may generate multiple snapshots. Within the current time range and the set of search hits, the trajectory list selects the newer snapshot of each trajectory, and does not combine the nodes of old and new snapshots into a complete execution.

Therefore, take note of the following items:

The number of trajectories in the list may differ from the number of records in Log View. A trajectory may correspond to multiple snapshot logs.
If the search conditions match only an old snapshot, a hit in the list does not necessarily mean that the latest status of the trajectory still meets the conditions.
When you open the details of a regular trajectory, the data is queried again. The details may be newer than the snapshot that is displayed in the list. To troubleshoot an exact historical status, check the record time and the processing version.

The page status and end time alone cannot prove business success. Make a judgment based on the actual inputs and outputs, the tool results, and the records on the business side.