Deployment and Upgrade of Coke Marketing Content

Data supporting coke-related marketing content mainly comes from domestic bulk commodity trading platforms, steel mill purchase ledgers, and port

What the data for this category looks like

Data supporting coke-related marketing content mainly comes from domestic bulk commodity trading platforms, steel mill purchase ledgers, and port stockyard inventory statistics. Data update rhythms are divided by dimension: spot transaction prices are updated daily, total port inventory is updated weekly, and industry supply and demand analysis reports are updated monthly. Most data is in structured tables or API return fields. Fields include delivery grade, sulfur content, ash content, volatile matter, origin, daily transaction price, port stock, and others. Units include yuan/ton, percentage, ten thousand tons, and more.

What constraints these characteristics impose on deployment and upgrade

Uneven data update rhythms, diverse field units, and large differences in document length create multiple constraints for deployment and upgrade. Different update frequencies of multi-source data require differentiated scheduled synchronization tasks, to avoid synchronization conflicts or data lag. Unit differences in structured fields require standardized mapping rules configured during the knowledge base parsing phase, to ensure accurate field matching during retrieval. Long-cycle industry report documents have large file sizes, requiring adjustments to text segmentation parameter thresholds to avoid content truncation during parsing. Meanwhile, marketing content has high timeliness requirements. Compatible configurations of original synchronization links must be retained during upgrades, to avoid service interruptions.

How to set configurations

Configuration ItemRecommended ValueRationale
PARSE_FILE_TIMEOUT_SECONDS600 secondsCoke industry monthly reports are mostly long texts, requiring sufficient parsing time to avoid mid-process timeouts
UPLOAD_FILE_MAX_SIZE2000 MBBulk inventory statistics CSVs and large supply and demand report PDFs have large sizes, needing adaptation to large file upload requirements
maxContext8000–12000 charactersCoke data includes multi-dimensional structured fields, requiring sufficient context to fully transmit field mapping rules
Recall CountTop 8 entriesMarketing content needs to cover multi-dimensional information including spot, inventory, supply and demand, with excessive entries causing content redundancy
Similarity Threshold0.75–0.85Need to filter low-match results from non-coke categories, while retaining valid data of the same grade but different origins
SYNC_CRON_EXPRConfigured per scenarioSpot data is synchronized every 15 minutes, port inventory is synchronized daily, industry reports are synchronized weekly

The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require specific analysis, and it is recommended to test on your own samples before finalizing settings.

Three common errors

  • A Request Timeout error occurs during conversation. The cause is failure to adjust the PARSE_FILE_TIMEOUT_SECONDS parameter for coke long documents, leading to knowledge base parsing timeout.
  • Tool functions cannot be called to complete data queries. The cause is incorrect configuration of SYNC_CRON_EXPR, leading to coke real-time spot data not being synchronized to the knowledge base, and the tool cannot obtain valid query parameters.
  • Docker image build fails, with logs showing file verification errors. The cause is failure to adjust the UPLOAD_FILE_MAX_SIZE parameter in the build configuration, leading to large coke industry reports being unable to be uploaded normally to the build environment.

How to confirm successful configuration

  • Upload a coke monthly report document, check if the parsed text retains all chapter contents completely, with no abnormal truncation.
  • Manually trigger a data source synchronization task, verify if the latest data of the corresponding category is successfully synchronized to the knowledge base.
  • Initiate a query containing multi-dimensional coke information, check if the returned results cover the preset field dimensions.
  • Test calling tool functions to complete data queries, confirm there are no timeout or missing parameter error messages.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.