Pick this sprint's fixes
Turn the recommendation list into a short, checked list of fixes with owners.
When to use: at sprint planning, to choose which recommendations your team takes on.
1. Clear the questions waiting on you
Type:
List my Single Origin recommendations. Which ones are waiting on us?
The agent calls list_recommendations. APPLIED_UNCONFIRMED ones sort first: Single Origin saw the change and the cost fell, and needs a person to say whether it was this recommendation. For each, the agent can call recommendation_outcome to read the measured saving and its margin.
Open Recommendations and pick the Did this work? tab. Answer each card. See Did this work?.
2. List the open recommendations by savings
Type:
Show me the top 10 open Single Origin recommendations by estimated daily savings.
The agent calls list_recommendations with statuses=["ACTIVE"]. Each row has title, type, estimated_daily_savings and query_signatures. Ask for page=1 for the next 10.
Pick the Open tab. Read Open savings at the top. Set Min Savings to hide small ones, and use the problem filter to focus on one kind.
3. Read each candidate
Type:
Walk me through recommendation 9c2b1f7e. Does the problem really appear in its query?
The agent calls get_recommendation for the problem (target_inefficiency and its SQL snippet), the fix and estimated_savings_summary. To check the problem is really there, it can call query_summary_for_job on the job named in query_signatures, then query_detail on a sample_query_id.
Click a row for the Quick look panel, then See the full recommendation for the cost chart, Problem, Fix, How the savings are estimated and the queries it covers.
4. Record your decisions
Type:
Move 9c2b1f7e to in review. Ignore 1de83a4c: "the table is being retired next month".
The agent calls update_recommendation with the recommendation's processing_time and status="IN_REVIEW", or status="IGNORED" with status_change_reason. A reason is required for IGNORED and INVALIDATED, and Single Origin carries it into the next analysis of the workload.
Use Actions → Mark as In Review, or Ignore with a reason. Use the assignee menu next to Status to name the owner. Moving it to In review assigns you when nobody is assigned yet.
A good final answer
This sprint: 3 fixes worth about 96 a day in estimated savings. (1) "Read only the last 3 days of analytics.orders_fact" (48.2/day), problem confirmed in its INSERT; owner Dana, in review. (2) Cluster key on analytics.events_raw (31.0/day). (3) Merge two duplicate daily rollups (16.8/day). Ignored 1 (table being retired). 2 applied recommendations confirmed on Did this work?.
Watch out for
- Estimates are estimates.
estimated_daily_savingsis projected; the measured saving comes after the fix ships, on Did this work?.- Fresh lists.
list_recommendationsresults are valid for one hour; ask again in a long session.- Overlap. Two recommendations on the same job or table may save less together than their sum.
- Status is the only thing you can change.
update_recommendationcan't edit a recommendation's text or savings.
Related
Updated about 7 hours ago