- Published
- Oct 31, 2026
- Updated
- Nov 19, 2026
- Reading time
- 23 min read
GA4 BigQuery Export Limit: Why Streaming Isn't Equivalent
Plan around GA4's daily BigQuery export limit. Audit streaming gaps, attribution fields, event exclusions and downstream completeness before changing modes.

In this guide (10 sections)
Streaming removes the daily export's event-number limit, but it is not a completeness-equivalent replacement. Standard GA4 daily exports have a one-million-event limit. Before switching modes or excluding events, define which raw events, attribution fields, reporting dates and completeness evidence your downstream analysis actually requires.
If daily export is paused, preserve the affected interval and existing tables immediately. Do not assume that enabling another mode, reducing tomorrow's volume or moving the rows to another warehouse will reconstruct missing past exports.
Real questions expose the tradeoff#
A Stack Overflow question about approaching one million exported events asks how to preserve traffic-source and channel information while increasing volume, eventually moving the data into another warehouse.
An independent question about combining exports across properties raises the same capacity problem. Replies propose streaming-only operation and make assumptions about intraday table persistence. One explicitly describes its retention explanation as a deduction from observed behavior.
These are older evergreen originals, not evidence of a current platform outage. They support one practical task: choose an export contract that survives the volume and still supports the business analysis. A workaround that delivers more rows but changes the required fields is a different dataset.
Document what your export needs to preserve
Use the checklist to compare required events, attribution fields and reporting dates.
Use the tracking QA checklistOpens a practical guide. No email required.
1. Confirm which limit you are approaching#
Record the property, service level, included streams, excluded event names, export modes and property timezone. Save any warning or pause notice and the last date for which the intended daily table actually exists. Inspect the selected project and dataset rather than inferring export health from a dashboard connection.
Google's export setup documentation scopes the million-event limit to Standard daily batch exports. It says persistent excess pauses daily export, prior days are not reprocessed, and sufficiently large excess can cause an immediate pause. It also documents stream and event exclusions as export-volume controls.
That is not a guarantee that the first million rows of every oversized day will arrive. Nor is a warning a universal grace-period promise copied from a forum. Read the actual account notice and retain its dates alongside the observed dataset state.
Separate export limits from collection, report sampling, storage quotas and dataset expiration. If the symptom is older history missing across different stores, use the GA4 history and retention audit. Increasing retention is not the same operation as resolving an export-volume pause.
2. Write down what must survive the change#
Inventory the downstream jobs before choosing an event to omit. A purchase dashboard, ordered funnel, engagement analysis and acquisition model can need different records.
| Requirement | Evidence to retain |
|---|---|
| Population | Included property, streams and event classes |
| Event detail | Required identifiers, parameters and supported schema fields |
| Time | Property day, event dates and extraction or arrival time |
| Attribution | Exact field path and whether it is collected or attributed context |
| Maturity | Provisional versus closed-day data and later revisions |
| Destination | Source table, load interval and downstream acceptance result |
| Completeness | Missing, deliberately excluded, not yet processed or unknown |
“Complete attribution” is not a field definition. Specify whether the analysis needs first-user acquisition, collected campaign information, session attribution or an aggregate UI channel report. A similarly named column in a different record does not establish equivalence.
Preserve the raw value and field path before applying a warehouse channel rule. If a channel is derived downstream, version that rule and identify its inputs. Do not label a custom classification as Google's complete reporting output merely because a total looks similar.
3. Compare freshness and completeness separately#
Google's BigQuery export reference describes streaming as best effort, without a completeness service-level objective, and with possible gaps. Its absence of an event-number cap does not guarantee every event arrived.
The same reference's mode table warns about new-user and new-session traffic-source availability. Its detailed streaming section names absent new-user fields: traffic_source.name, traffic_source.source and traffic_source.medium. It also recommends the daily export for user-attribution analysis rather than relying on streaming while those fields are processed.
Keep that statement at its documented scope. It does not mean every collected campaign parameter is always absent, or that every field called source is interchangeable. Inspect the actual schema and records needed for the question.
The export schema reference separately describes collected_traffic_source, session_traffic_source_last_click and traffic_source. Their names reflect different contexts. Replacing a missing acquisition value with another available field may create a new analytical assumption; record it rather than silently presenting it as recovered data.
| Candidate response | What it changes | What still needs proof |
|---|---|---|
| Reduce exported streams or events | The selected daily-export population | Required analyses still have their inputs |
| Add or use streaming | Recent-event availability and mode contract | Gaps, required fields and downstream maturity |
| Evaluate higher service-level limits | Available product capacity | Actual account eligibility, cost and operating requirements |
| Use a reporting API for a summary | The source becomes report data | It satisfies the summary task, not a raw-log reconstruction |
Treat each as an owner-reviewed design choice. There is no single switch that simultaneously proves volume coverage, field availability, low latency and parity with Analytics reports.
Diagnose a reported streaming gap at the boundary that owns it
A recent Analytics Community question about apparently missing streaming-export events describes a user-reported difference between GA4 reporting and BigQuery. The post's figures and proposed cause have not been independently verified. The useful response is to locate the boundary, not to assume that every difference is an export loss.
First record the property reporting time zone, the event date being compared, query time, table suffix, stream, event names, filters, and whether the interval is complete. For the current day, events_intraday_YYYYMMDD is a provisional table: Google says it can omit records because of late-event processing or failed uploads and is deleted after the completed events_YYYYMMDD table is ready. Google also publishes typical data-freshness timings, explicitly without an SLA; some data can arrive late. Do not compare a partial intraday table to a settled report and call the difference permanent. Use the BigQuery export guide and Analytics freshness guidance for the current property and export mode.
Next, separate three populations:
| Observation | What it can establish | What it cannot establish alone |
|---|---|---|
| GA4 standard report or Explore | The selected report's processed metric, identity, filters and reporting settings | A row-for-row raw event count in BigQuery |
Finalized events_YYYYMMDD table | Rows available in that completed export table under the property's export contract | That every UI metric or modeled value must match raw rows |
events_intraday_YYYYMMDD table | Provisional current-day rows that have reached the streaming export | A completed-day total or a completeness guarantee |
| A downstream load or transformation | Its own inputs, job status, filters and output | That an upstream GA4 export delivered every intended event |
Use matched dimensions and units before subtraction. A report's sessions, modeled users, key events or filtered event count are not interchangeable with BigQuery event rows. Check the reporting identity and any report filters, and inspect the query for date boundaries, event-name selection, duplicated wildcard tables, joins, UNNEST row expansion and downstream exclusions. A query that counts parameter rows after UNNEST can count a different unit than one row per GA4 event.
If you operate a separate writer that streams rows into BigQuery, the BigQuery STREAMING_TIMELINE view can show per-minute requests, rows, bytes and errors for REST streaming writes in the specified project and region. It is telemetry for that BigQuery ingestion path; it does not expose GA4's collection endpoint, and its presence or absence does not by itself prove whether the GA4 export delivered an event. Confirm that the table and write method under investigation are actually represented before using the view as evidence.
Only after comparing completed days, matching definitions and checking your own downstream pipeline should you escalate a suspected export issue with the affected dates, table states, query, export configuration and reproducible counts. Keep a reported discrepancy separate from a confirmed missing event, and do not invent a percentage or a service-level commitment from one property's observations.
When the daily table and export job stop appearing
Treat an absent daily table as a symptom to localize, not proof of a Google incident. A November 13 operator report describes multiple GA4 properties whose events_YYYYMMDD tables stopped after a date while streaming tables continued. The author reports checking links, billing, roles and project logs. Later commenters report similar symptoms; one quotes a response from their Firebase Support case describing an engineering investigation, and several users later report that daily tables returned on different dates. These are community reports and a quoted support response, not a public Google incident notice independently verified for this guide. Treat them as a reason to preserve per-property evidence and ask support about status, not proof of a current outage, universal resolution, cause, or guaranteed backfill.
Start by naming the missing object and the evidence boundary. For every affected property, record its GA4 property ID, linked project and dataset, export mode, property time zone, expected table suffix, last completed events_YYYYMMDD table, current events_intraday_YYYYMMDD tables, and the time at which each was checked. Record the project and region used for BigQuery job history. A table that is not yet expected under the property's day boundary is different from a closed day that has no table.
Then inspect the export path in this order:
- In GA4, confirm that the intended property remains linked to the expected Cloud project and dataset, that the daily export mode is enabled, and that neither selected streams nor excluded event names explain the population. Save the configuration and change-history evidence.
- In the destination project, check dataset existence and location, billing state, quotas, service identity and relevant IAM changes. These are checks, not presumptive explanations.
- In BigQuery job history, filter the affected interval and dataset for the export job. Distinguish no matching job, a job that failed, and a successful job whose output is not where expected. Preserve the job ID, state, error reason, location and creation time for any job that exists. “No job found” is evidence only for the selected project, region, interval and filters; verify those filters before escalating.
- Compare a second property or project only after normalizing the reporting date and table naming. Parallel reports strengthen the case that the symptom spans accounts, but do not identify a cause.
- If streaming is enabled and its intraday table receives rows, treat that only as evidence that this separate path is producing provisional data. Google's BigQuery export guide says streaming export is best effort and not a completeness guarantee. Do not sum intraday and final daily tables or present streaming rows as a restored daily batch.
Google's export setup guide describes daily export configuration and service limits. Its data-freshness guidance explains that reporting data can take time to process and gives typical delays rather than a completion promise. Those timing notes help distinguish a delayed result from a missing attempt; they do not explain a persistent absence of jobs and are not an incident confirmation.
Avoid toggling export modes, unlinking the property, changing exclusions, or rebuilding data from streaming before preserving the original state and the affected interval. Google documents that historical data from before an export setup or change is not necessarily reconstructed; treat the availability of any missing past day as unknown until the source confirms it. If you contact support, provide the property/project identifiers privately, exact date boundaries, job-history filters, table inventory, configuration evidence and the first missing day. Do not publish account identifiers in a public thread.
Distinguish delayed, failed and absent exports
| Observation | What it supports | What remains unknown |
|---|---|---|
| Daily table has not appeared yet and the property day is recent | The day may still be processing | Final arrival time or completeness |
| Export job exists with a failed state | A job was attempted and failed at the recorded boundary | Whether a retry will occur or how many days are affected |
| No matching job appears after checking project, region and time filters | No matching attempt is visible in that query | Whether the wrong project/region/filter was used or why the attempt is absent |
| Intraday table receives rows while daily table is absent | The streaming path is producing provisional rows | Daily batch completion, full event coverage or historical recovery |
| Daily table exists but counts differ from an Analytics report | Two selected populations or processing paths differ | A data-loss diagnosis until definitions and completion state match |
Keep the observation time beside every state. Recheck the same project and date boundary later, append the result instead of overwriting it, and separate verified tables/jobs from the user's reported symptoms. If one day appears later, that establishes its observed arrival time only; it does not guarantee that other properties or dates will resolve identically.
4. Use deliberate exclusions, not invisible loss#
List event volumes over representative completed days, including known peaks. Ask which analyses depend on each event before proposing export exclusions. The fact that an event is common does not make it disposable: it may supply required journey, engagement or quality evidence.
Changing the BigQuery export selection is a different boundary from changing the site's collection. Record both configurations so a future analyst knows whether an event was not collected, excluded from export or absent for another reason.
Do not exclude the events required to validate the change and then claim the remaining dataset proves complete measurement. Keep a versioned export-scope manifest, the approval, effective time and any affected dashboard definitions. Make the reduced population visible in documentation and downstream labels.
Hypothetical volume plan
Suppose an invented Standard property has the following completed-day event counts in a trusted planning inventory:
| Event category | Event occurrences |
|---|---|
| Pageviews | 450,000 |
| Scroll activity | 250,000 |
| Engagement activity | 300,000 |
| Purchases and other required events | 200,000 |
| Total | 1,200,000 |
Assume the analysis owner separately confirms that this particular export's intended tasks do not need the scroll category. Excluding those 250,000 events would leave 950,000 selected event occurrences, or 79.2% of the original 1,200,000 after rounding. The 250,000 excluded events are deliberate scope reduction, not lost collection and not a measured streaming failure.
The selected volume leaves 50,000 events below the million-event daily limit for this hypothetical day. That is 5% of the limit, not protection against every future spike. Review peak days and changing event mix before accepting the plan.
No real property's data is represented here. The example does not recommend dropping scroll events generally, and it does not prove that a paused earlier day will be exported after the change.
5. Build a maturity rule for downstream loads#
Keep provisional current-day ingestion separate from the closed-day source used for final reporting. The schema documentation identifies daily and intraday table families and explains the intraday table's removal when the daily table is complete. Avoid blindly adding both versions of the same day's activity together.
For each downstream load, retain the table names, extraction time, selected dates, row counts, schema checks and event-scope manifest. If records can arrive again in a later extraction, define an appropriate replacement or reconciliation policy with the data owner. Do not invent a universal deduplication key from event name and timestamp alone; legitimate events can share those values.
Late-arrival handling is a separate concern from a paused export. Google's references describe normal updates to recent daily tables, and the main export guide also notes occasional historical reprocessing. Do not convert those descriptions into a blanket exact-hour completion guarantee or a promise to reconstruct days omitted after a volume pause.
If an event day is incomplete, preserve its status as incomplete. A zero-row downstream result could mean a genuinely empty selected population, an absent source table, an unready load or a failed job. Give those states different labels and escalation paths.
An additional warehouse cannot restore an event that was never available in the source dataset. Moving or unioning tables requires its own scope, overlap and identity review; a single destination table does not prove a unified reporting history.
6. Do not assume a 360 feature belongs to a Standard property#
Google's 360 feature comparison documents higher export limits. Its Fresh Daily guide scopes that mode to eligible Normal and Large 360 properties and describes its availability and limitations. Do not present it as a free Standard setting or a purchase recommendation without reviewing the actual requirement.
The main export guide documents a completeness signal for Daily or Fresh Daily exports with 360 properties. Confirm the mode and log scope; do not promise that the same signal exists for a Standard export. An operational signal is also not evidence that every downstream transformation and destination load completed correctly.
For a candidate design, demonstrate the required schema and maturity on a representative supported interval. Compare like-for-like dates and scope, document unresolved fields and ask the product or data owner to approve the acceptance criteria. A vendor's example tolerance is not a universal completeness standard.
7. Keep report extracts separate from raw-event alternatives#
Google's Data API overview describes programmatic report access governed by the property's reporting identity. An API summary can be useful when the business only needs a defined aggregate. It is not a replacement source for unavailable individual raw events.
A report-API alternative also has its own capacity contract. Identify GA4 quota errors, measured token usage and concurrent-request pressure before relying on a scheduled summary. Preserve the intended report population and a dated fallback; fewer returned rows or a warehouse migration does not by itself prove equivalent reporting.
Google's UI-versus-export explanation likewise distinguishes collected event data from reporting additions. Keep the exact metric and query when comparing a warehouse result with the interface. Do not manufacture rows from aggregate counts or assign modeled totals to individual records.
The platform measurement comparison provides a broader contract for those comparisons. When the initial issue is absent incoming events, the GA4 receiving-path investigation owns that earlier boundary. A successful export mode cannot repair a website event that was never sent through the intended collection path.
Keep a useful release and incident record#
Retain the warning, selected event volumes, affected dates, original export settings, proposed scope, required fields and downstream job evidence. State which historical tables exist and which intervals remain unavailable. Recheck the approved design after a release or material change in volume; do not turn one successful day into a permanent completeness guarantee.
For an AdProof-measured campaign, export its supported dates, timezone and available dimensions as a separately defined delivery dataset. The anonymized fashion-retail delivery-quality case illustrates an ad-impression baseline, not GA4 export recovery. AdProof's campaign-export workflow does not raise Google's limits, reconstruct paused BigQuery exports or provide an assumed direct warehouse integration. Discuss a scoped delivery review with each dataset's event unit and coverage explicit.
When a daily table or report date disappears, isolate the boundary first
An absent events_YYYYMMDD table, an existing table with fewer rows than expected, a report day missing in the GA4 interface, and a downstream query returning no rows are different symptoms. An events_intraday_YYYYMMDD table does not establish that the daily batch has completed or that a report is fully processed. Start by recording the property timezone, affected dates, last successful daily table, table existence and row counts, export mode, and the latest time each source was checked.
- Confirm dates and table family. Daily table suffixes follow the Analytics property timezone. Check whether the date is still within Google's documented late-event update period, and do not add intraday and daily tables for the same date as if they represented disjoint event populations. Google says the daily table can be updated for up to two calendar days plus today for late-arriving events; this is not a universal promise that a missing table will arrive at an exact hour.
- Check whether export was paused or not attempted. Review export setup and any status or limit notice, the BigQuery dataset, job history and permitted logs. A failed load, no export attempt, a paused daily export, and an empty completed table point to different owners. Standard daily export has a one-million-event limit; do not infer the limit caused a stop when the affected property is below it or the job was never attempted.
- Check linked-project prerequisites without assuming the banner is the cause. Google says export requires a valid Cloud payment method. Verify billing state and permissions from the authorized Cloud project owner's account. A transient or ambiguous “could not retrieve billing status” message is evidence to investigate, not proof of a payment failure or a Google-wide billing defect. Preserve the exact message and account context, then ask Google support to confirm the export-side state if the checks disagree.
- Compare GA4 reports and BigQuery using the same event question. A missing report date can reflect report processing or a selected property/date boundary; a missing table is an export symptom. Compare the same property timezone, date and event family. Sessions, modeled report metrics, exported event rows and downstream transformed counts are not interchangeable.
- Preserve current evidence before switching modes. If streaming is already enabled, retain the intraday table at an agreed checkpoint because Google removes it after the daily table is complete. Streaming is best-effort and has its own field and completeness contract. Turning it on can provide a separate interim dataset, but it does not retroactively reconstruct a failed daily export or make the two modes equivalent.
- Escalate a reproducible gap. Send the affected property, project and dataset identifiers through the authorized private support channel, dates/timezone, export mode, expected/observed table state, job-history evidence, last known successful export and exact error text. Request confirmation of whether the missed dates are expected to be re-exported; do not promise backfill until the actual tables show it.
| Observation | Next check | Do not conclude yet |
|---|---|---|
| Intraday table exists; daily table is pending | Compare property timezone, expected daily freshness, late arrivals and processing | Daily completeness or a recoverable historical gap |
| No daily table and no recent export job attempt | Linked configuration, export status, project and support-side export state | That an event-volume cap or tag failure caused it |
| Job ran and failed | Exact job error, permissions, billing and destination state | That all same-date data is permanently absent |
| Daily table exists with fewer rows | Event filters, schema, collection time, late-event window, query and downstream transforms | That every difference is upstream GA4 loss |
| GA4 report day is missing while export is present | Report identity, filters, property timezone and report freshness | That the table is incomplete or the report must match raw rows |
In the Nov 11–14, 2026 discussion cluster, posters reported missing daily tables and separate report/Google Ads processing symptoms; a support reply was quoted by a poster, while later replies contradicted an early billing explanation. The public Google status history, checked on 14 November 2026, did not show a matching GA4 export incident. Google's current export documentation, setup guidance, and schema reference define the stable troubleshooting boundaries.
Validate recovery after a delayed daily table returns
A table's arrival is one recovery signal, not proof that every downstream output is complete. In a Nov 15 follow-up to the Nov 13 report of missing daily exports, the author said daily tables had returned sequentially and some contained more rows than earlier intraday snapshots, while the newest report date still looked abnormal. Separate Analytics Community reports describe missing regular tables while _fresh tables were present and a date that appeared in a report before returning to zero. These are user reports, not an official service-wide resolution or a promise of backfill.
Record recovery as versioned checkpoints:
| Checkpoint | Capture | Interpretation limit |
|---|---|---|
| Daily source | Table/date, first-seen time, property timezone, event and stream counts | Existence does not establish report or downstream maturity |
| Earlier intraday snapshot | Capture time, date, filters and event counts | Provisional snapshot may differ from later daily data |
| Warehouse transformation | Source table/version, query version, job time and output | A completed query may still use an earlier source snapshot |
| GA4 report | Metric, dimensions, filters, property timezone and refresh time | Report metrics such as sessions are not raw BigQuery event-row counts |
| Ads conversion view | Action, date basis, attribution window and extraction time | A table arrival does not prove an imported conversion has settled |
Compare like with like: the same property-local event date, stream, event filter and scope. Daily tables can receive late-arriving events and may be reprocessed, so a higher row count than an earlier intraday copy is not by itself evidence of duplicate data. Conversely, do not add daily and intraday records for the same date into one final population. Keep a source-version manifest with extraction time, query version and row/event definitions. If a downstream table was materialized before the daily table appeared, verify whether the normal pipeline replaced or reprocessed the affected date instead of silently appending another copy.
After the source table arrives, rebuild only the affected downstream window through the established idempotent process and retain the previous output for comparison. Then recheck the GA4 report and any Ads import on their respective date bases and processing windows. State which surfaces are verified current and which remain pending; table creation, report display, attribution credit and business outcome are distinct observations.
In an explicitly hypothetical example, an intraday snapshot contains 9,800 event rows for a property-local date and the later daily table contains 10,050. The 250-row difference is not automatically duplicate or missing traffic. Compare event names, streams, dates, filters and transformation versions. If the final daily source has 10,050 rows but a downstream report still reflects the earlier snapshot, that supports a stale-source investigation; it does not explain why an unrelated interface report may differ.
Frequently asked questions#
Can I avoid the daily event limit by choosing streaming?
Streaming has no event-number cap in the documented export setup, but its completeness and field-availability contract differs. Validate the analysis requirements rather than treating additional rows as equivalent to a complete daily export.
Will excluded events stop appearing everywhere in GA4?
An export exclusion concerns the selected BigQuery dataset. Keep it distinct from changes to collection or other report settings. Document the effective export scope so downstream users do not mistake deliberately omitted events for a technical failure.
Will reducing volume recover days missed during a daily-export pause?
Do not assume recovery. The setup documentation says previous days are not reprocessed after the stated export-limit pause. Inventory actual surviving tables and keep that limitation separate from ordinary recent-table updates or exceptional historical reprocessing.
Is a report API or another warehouse a complete raw-log substitute?
No automatic equivalence follows. A reporting API supplies report data, while another warehouse can hold only the records actually transferred under its source contract. Verify event detail, field availability, scope and maturity before treating either as the requested raw history.
Document what your export needs to preserve
Use the checklist to compare required events, attribution fields and reporting dates.
Use the tracking QA checklistOpens a practical guide. No email required.
