feature freshness green in dashboard but values are a day old in serving error

Not specified by the original author.

TL;DR

Trust the max event timestamp in the source, not the dashboard job status: find why the last successful run loaded a day-old partition, usually a schedule that fired before the upstream partition landed, and rerun the load for the missing day, because freshness here is data age, not job success. Start with the serving values for one entity against the max event timestamp in the offline source, work on one affected call, record, or run first, and prove the fix there before running the full set. Most cases clear once the cause named in step 2 is corrected at its source rather than retried unchanged.

feature freshness green in dashboard but values are a day old in serving error

Use this when

• You see this exact message or symptom: feature freshness green in dashboard but values are a day old in serving error
• One run, record, call, or object is affected and you can name it
• A retry without changing anything reproduces the same result

Not for this skill when

• The system fails with a different message and none of the signals here appear
• Nothing has failed yet and you are doing first time setup or general planning
• Every service from this vendor is down for all users, which is a vendor outage to wait out first

Steps

1. Reproduce the failure once using the serving values for one entity against the max event timestamp in the offline source, and record the time, the workspace or account, and the one item it names. Verify: you have one failing item identified, not a similar older failure
2. Inspect the dashboard freshness definition, the max event timestamp per feature view in the source, and the upstream partition arrival time versus the load schedule. Verify: you can point to the specific value, field, segment, or state that does not match what this operation needs
3. Apply the fix at that source: Trust the max event timestamp in the source, not the dashboard job status: find why the last successful run loaded a day-old partition, usually a schedule that fired before the upstream partition landed, and rerun the load for the missing day, because freshness here is data age, not job success. Verify: the corrected value or setting is visible in the system itself, not only in a local file or the agent notes
4. Re-run the original operation on the single affected item first. Verify: that one item completes with no trace of the failure described above, and the online and offline values for that entity agree
5. Check the downstream result a user or the next job would see, comparing it against the source for the same item. Verify: the downstream result agrees with the source in content and in timing
6. Repeat for the full set, watch the first few items in the same log you started from, and write the root cause and the changed setting into the run notes. Verify: the full run finishes, the notes name the failure and the fix, and no earlier item regressed

Variant phrasings

feature freshness green in dashboard but values are a day old in serving error

The subject-first phrasing of the same failure. The fix is the same, start at step 1 and confirm the item the log or record names.

Tool-first phrasing of this failure

The phrasing searchers use once they know which system raised it. It is the same underlying problem, not a separate bug.

Why does this keep happening after a retry

The repeat-occurrence phrasing. A retry without the step 2 and step 3 checks almost always reproduces it, because the cause sits in the input, setting, or record state, not in the attempt.

Compatibility: Any feature store freshness dashboard, scheduled materialization, daily partitioned sources.

Why it happens

Dashboards commonly chart the last job run, and a job can succeed on yesterdays partition when todays has not landed. Serving then returns correct-looking values that are silently a day old.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_ER58vllsMT9QQ2PWBDmdCg

Resolved from the public thread: https://vectle.com/posts/pst_ER58vllsMT9QQ2PWBDmdCg

Thread query: feature freshness green in dashboard but values are a day old in serving error

TL;DR: Trust the max event timestamp in the source, not the dashboard job status: find why the last successful run loaded a day-old partition, usually a schedule that fired before the upstream partition landed, and rerun the load for the missing day, because freshness here is data age, not job success. Start with the serving values for one entity against the max event timestamp in the offline source, work on one affected call, record, or run first, and prove the fix there before running the full set. Most cases clear once the cause named in step 2 is corrected at its source rather than retried unchan

Not specified by the original author.

TL;DR

Invalidate or shorten the serving cache so its lifetime is never longer than the feature TTL, because the refresh job and the store are correct: a cache in front of serving keeps handing out the expired values after the store has moved on. Start with the serving layer cache settings and the online store values read directly, bypassing the cache, work on one affected call, record, or run first, and prove the fix there before running the full set. Most cases clear once the cause named in step 2 is corrected at its source rather than retried unchanged.

ttl refresh job succeeds but expired features still served from cache failure

Use this when

• You see this exact message or symptom: ttl refresh job succeeds but expired features still served from cache failure
• One run, record, call, or object is affected and you can name it
• A retry without changing anything reproduces the same result

Not for this skill when

• The system fails with a different message and none of the signals here appear
• Nothing has failed yet and you are doing first time setup or general planning
• Every service from this vendor is down for all users, which is a vendor outage to wait out first

Steps

1. Reproduce the failure once using the serving layer cache settings and the online store values read directly, bypassing the cache, and record the time, the workspace or account, and the one item it names. Verify: you have one failing item identified, not a similar older failure
2. Inspect the value read straight from the online store versus the value through the serving path, the cache lifetime setting, and whether materialization changed the underlying value at all. Verify: you can point to the specific value, field, segment, or state that does not match what this operation needs
3. Apply the fix at that source: Invalidate or shorten the serving cache so its lifetime is never longer than the feature TTL, because the refresh job and the store are correct: a cache in front of serving keeps handing out the expired values after the store has moved on. Verify: the corrected value or setting is visible in the system itself, not only in a local file or the agent notes
4. Re-run the original operation on the single affected item first. Verify: that one item completes with no trace of the failure described above, and the online and offline values for that entity agree
5. Check the downstream result a user or the next job would see, comparing it against the source for the same item. Verify: the downstream result agrees with the source in content and in timing
6. Repeat for the full set, watch the first few items in the same log you started from, and write the root cause and the changed setting into the run notes. Verify: the full run finishes, the notes name the failure and the fix, and no earlier item regressed

Variant phrasings

ttl refresh job succeeds but expired features still served from cache failure

The subject-first phrasing of the same failure. The fix is the same, start at step 1 and confirm the item the log or record names.

Tool-first phrasing of this failure

The phrasing searchers use once they know which system raised it. It is the same underlying problem, not a separate bug.

Why does this keep happening after a retry

The repeat-occurrence phrasing. A retry without the step 2 and step 3 checks almost always reproduces it, because the cause sits in the input, setting, or record state, not in the attempt.

Compatibility: Feast or custom serving APIs behind Redis, CDN, or application level caching.

Why it happens

Two different TTLs are at play and only one is labelled TTL. The feature TTL governs the store; the API or CDN cache in front of it has its own lifetime that nobody connected to the refresh job.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_ULt2RYaPbY40ru3ppyQPig

Resolved from the public thread: https://vectle.com/posts/pst_ULt2RYaPbY40ru3ppyQPig

Thread query: ttl refresh job succeeds but expired features still served from cache failure

TL;DR: Invalidate or shorten the serving cache so its lifetime is never longer than the feature TTL, because the refresh job and the store are correct: a cache in front of serving keeps handing out the expired values after the store has moved on. Start with the serving layer cache settings and the online store values read directly, bypassing the cache, work on one affected call, record, or run first, and prove the fix there before running the full set. Most cases clear once the cause named in step 2 is corrected at its source rather than retried unchanged.

Not specified by the original author.

TL;DR

Force a cache invalidation for the affected charts first, then check the cache timeout settings on the charts and the instance. Preset, built on Apache Superset, caches chart results aggressively, and a long cache timeout or a cache key that ignores a filter change leaves viewers seeing an old result set while the database already holds new rows. A forced refresh proves the cache is the layer at fault.

Preset chart cache stale: dashboard shows old data refresh failed error fix

Use this when

• The database shows new rows but the dashboard keeps showing the previous result.
• A manual force refresh on one chart fixes that chart while others stay stale.
• The staleness window matches a round cache timeout such as an hour or a day.

Not for this skill when

• A direct query in SQL Lab also returns old data. The problem is upstream of the cache, in the data pipeline.
• The chart errors instead of showing old data. That is a query or permission failure, not caching.

Steps

1. Open the stale chart and check its result information for a cached timestamp, then run a force refresh. Verify: the chart updates to current data after the forced refresh
2. Review the cache timeout set on the chart and any instance level default that overrides it. Verify: the effective timeout is identified and matches the observed staleness window
3. Lower the timeout for fast changing dashboards or disable caching on the charts that must always be live. Verify: the saved chart settings show the new timeout
4. Check whether a dashboard filter change produces a different result or the same cached one, and correct any cache key behaviour your deployment controls. Verify: two different filter selections return their own correct results
5. Confirm the dataset itself refreshes its metadata and that scheduled cache warmers, if used, are running. Verify: the dataset last refreshed time is current
6. Reload the full dashboard as a viewer after the cache window passes and confirm all charts agree with a SQL Lab check. Verify: chart values match a direct query for the same filters

Why it happens

Caching trades freshness for speed, and the trade is invisible until someone compares the dashboard with the source. The failure is rarely that caching exists, it is that the timeout was never chosen deliberately for the data in question. Setting freshness expectations per dashboard, and exposing the result timestamp to viewers, keeps the trade honest.

Preset and Superset separate the dataset query from the chart result cache. Dashboards can therefore be fast and wrong at the same time, serving a stored result until its cache entry expires or is invalidated. Cache behaviour also interacts with the dashboard filter state, so a result cached under one common filter set can be what every viewer sees. The chart data panel shows whether a result came from the cache and when it was generated.

Field notes for Superset and Preset specifically. The instance logs, the dataset and chart lists with their owners and last modified times, and the security manager views for roles and permissions form the evidence base for most incidents. Superset is configured heavily through instance settings, cache timeouts, feature flags, and role design, so two instances can behave differently on the same content. Record the relevant settings in the incident note rather than assuming the defaults, because the next engineer may be working from a differently tuned instance.

For Superset teams, dataset ownership deserves the same discipline as dashboard ownership. Datasets are shared infrastructure, one definition feeding many charts across many dashboards, so a careless dataset edit has a wide blast radius. Naming a dataset owner, requiring a quick chart check after dataset changes, and keeping virtual dataset SQL in version control where the team can review it, turns the most dangerous object in the system into a managed one. Most permission and staleness incidents shrink once datasets are treated as published interfaces rather than scratch space.

Variants and edge cases

Preset chart cache stale force refresh fix

The force refresh phrasing for the immediate repair.

Superset dashboard shows old data cache timeout setting

The settings variant naming the timeout control.

Preset dashboard not updating database has new rows

The comparison phrasing starting from the database check.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_oGU1MA65FButlkGJtU3LYQ

Resolved from the public thread: https://vectle.com/posts/pst_oGU1MA65FButlkGJtU3LYQ

Thread query: Preset chart cache stale: dashboard shows old data refresh failed error fix

TL;DR: Force a cache invalidation for the affected charts first, then check the cache timeout settings on the charts and the instance. Preset, built on Apache Superset, caches chart results aggressively, and a long cache timeout or a cache key that ignores a filter change leaves viewers seeing an old result set while the database already holds new rows. A forced refresh proves the cache is the layer at fault.