TRACKING SPECIALIST — BARIS ASA · LONDON
Synctag
Start the check → TRACKING SPECIALIST — BARIS ASA · LONDON

READING ROOM › SYMPTOM FILES › SYM-025

SYMPTOM FILE SYM-025 FILED 05 NOV 2026 · REVISED 06 NOV 2026

Server-side GTM errors: what Preview shows and what to check.

ALSO FILED AS
  • “server-side GTM errors”
  • “who fixes server-side GTM errors”
  • “server-side GTM not receiving requests”
  • “fix server-side Google Tag Manager errors”
  • “sGTM preview not showing requests”
  • “server-side GTM preview not working”
  • “server container preview empty”
  • “server-side GTM tag failed”
  • “The request was aborted because there was no available instance”
  • “Cloud Run 429 server-side tagging”
  • “hire a server-side GTM expert UK”
  • “GA4 client not claiming request server-side GTM”
  • “server_container_url not working”
  • “server-side GTM losing data on Black Friday”
  • “how to debug server-side Google Tag Manager”

Baris Asa, tracking consultant, London · Last checked: November 2026

If GA4 or Google Ads have shown fewer purchases than your store since you moved to server-side GTM (your tags running on a server you pay for, not in the visitor’s browser), the loss sits in one of five places: the server is down, the site isn’t sending to it, the server ignores what arrives, a tag fails, or the server is full on sale days. Check them in that order.

Still mid-move? The staged plan, with a check after each step, is in moving from client-side to server-side GTM.

  1. The server is down. Its address with /healthy on the end doesn’t show the word ok. Check
  2. The site isn’t sending to it. Preview (GTM’s test mode) stays empty while you browse the site. Check
  3. The server ignores what arrives. Requests reach the server, but nothing in it picks them up. Check
  4. A tag fails. A purchase tag is missing from the tags that fired, or marked as failed. Check
  5. The server is full on sale days. The gap widens only at peak, and the server’s logs say “no available instance”. Check

Can’t check these yourself? Who should fix it, and what to ask for. If all five are clean, the last section covers double counting and Safari.

Who should fix server-side GTM errors, and what to ask them for?

Hire someone who reads server Preview before changing anything, follows one test order from the browser to each platform’s response, and runs the same test order again after the fix. They should be able to work on the server as well as the container: instance limits and logs on Google Cloud, or the host’s own logs on Stape or Addingwell. I’m Baris Asa, a London tracking consultant, and that’s the server-side tracking work I do for UK ecommerce brands and their agencies.

What a good fixer asks for, and shows you:

  • Access: publish rights on the server container, and on Google Cloud a role that can read logs and change instance settings. Removed when the job is done.
  • Evidence: the same test order in Preview, before and after the fix.
  • A cause in one of the five places, not “we rebuilt the container”.

Before you hire anyone, collect these three. They tell you, and whoever you hire, where to start.

  1. GA4 purchases next to store orders, day by day. Do this one yourself first. Take last year’s sale weekend and a normal week. If you moved after last year’s sale, compare your busiest day since the move with a normal day.

    • A small gap every day is normal: declined consent and ad blockers see to that.
    • A gap that jumps only on sale days points at server capacity (place 5).
    • A steady gap that started with the move points at places 1 to 4.

    Capacity changes have to go in before a sale. Run this before Black Friday.

  2. The server’s address with /healthy on the end, opened in a browser. A working server shows the word ok. An error page, or nothing, means the server isn’t answering.

  3. A screenshot of server Preview after one test order, on the Tags tab. Every purchase tag (GA4, Google Ads, Meta) should be listed under Tags Fired. A purchase tag marked as failed, or missing, is the fault. On Stape or Addingwell, ask the host for the same screenshot.

Send whichever of the three you have, plus your store platform, to [email protected]. I’ll tell you where it breaks before I quote, then fix it and rerun the same test order.

Where does server-side tagging break?

Server-side tagging runs your tracking tags on a server you control instead of in the visitor’s browser. The Google tag on your site (or another source) sends events to that server, and the server forwards them to GA4, Google Ads, Meta and the rest (Google: what is server-side tagging).

It adds to browser tagging rather than replacing it. In Google Tag Manager the server side is a server container (a set of tags and rules) on a server you pay to host. It doesn’t replace consent: Consent Mode v2 with server-side GTM. The full comparison: server-side vs web GTM.

Brands usually move to it for more control over what reaches each platform, and to recover some of the conversions lost to ad blockers and Safari’s cookie limits (F-401 ). How much it recovers depends on the domain set-up, covered at the end.

It also adds two links the browser-only set-up never had: the trip from the browser to your server, and the server itself. Those two links, plus the container’s wiring, are the five places.

Everything below is for whoever manages the containers.

How do I read server-side GTM Preview?

Open the server container in Tag Manager and click Preview (Google: preview and debug server containers). The left panel lists incoming requests, such as collect?v=2&.... Each event the request carried sits under it. On the right:

  • Request tab. Three boxes: the client that claimed the request, the incoming HTTP request, and the outgoing HTTP requests. The incoming request should match what Chrome’s Network tab shows the browser sent.
  • Tags tab. Tags Fired and Tags Not Fired. Each fired tag is marked as failed or successful, with how many times it fired.
  • Variables and Event Data. What the server built from the request. A value missing here is missing for any tag that reads it from the event.
  • Console. Errors from fired tags, each with its message and source.

One quirk before you diagnose: GA4 normally bundles several events into one request, but not while the web container is also in Preview. Different grouping in the two modes is normal.

Preview won’t open, or opens and shows nothing

What you see: the preview page doesn’t load, or it loads and no requests ever appear. Likely cause: the tagging server or the preview server isn’t reachable, or preview traffic can’t find its way back to you. First check: /healthy on the Server container URL from Admin, Container Settings. Google treats a loading preview page as the basic set-up check; if it loads but shows no events, carry on below (Google: Cloud Run set-up).

If /healthy fails on a new custom subdomain, check that the address resolves and answers over HTTPS before anything in GTM. The server runs on your own domain only once DNS (or your CDN or load balancer) points there (Google: custom domain). On a managed host such as Stape or Addingwell, use the host’s own status and set-up checks.

If the server is healthy and Preview is still empty, check how preview works. Opening Preview sets a preview cookie in your browser. Requests from that browser carry it, so the tagging server knows they’re yours and forwards them to the preview server named in its PREVIEW_SERVER_URL setting (Google: private preview server). So:

  • Browse the site in the same browser you opened Preview in. Another browser, or a private window, sends no cookie.
  • Match the two addresses. Preview opens from the Server container URL in Admin, Container Settings. If the web tag’s server_container_url is a different host (the custom domain, say, while Container Settings still holds the default one), in my experience the cookie doesn’t travel with the requests and Preview stays empty. Google doesn’t document the cookie’s domain; making the two match is the quick test.
  • Check PREVIEW_SERVER_URL on the tagging server points at the live preview server.
  • If you route the preview server through your load balancer (Google’s private preview set-up on Cloud Run), give it its own host, such as preview.sgtm.example.com. Mapping a path such as /gtm/debug at the load balancer shows the debug window but no events. In the default set-up, PREVIEW_SERVER_URL points straight at the preview service and the load balancer needs no preview rule.
  • On a manual (self-hosted) set-up: exactly one preview server, no autoscaling, and a load balancer or CDN timeout over 20 seconds, or preview won’t work properly. A same-origin preview server sits at a different path, with that path in PREVIEW_SERVER_URL (Google: manual set-up).

No request arrives at the server

What you see: Preview opens, you browse the site, the left panel stays empty. Chrome’s Network tab shows GA4 requests going to google-analytics.com, or nowhere. Likely cause: the browser isn’t sending to the server, or something on the page stops it. First check: the web container’s Google tag. It needs a server_container_url setting with the server’s address (Google: introduction to server-side tagging). A typo, an old default address, or a second Google tag without the setting all send the data elsewhere.

Then the page itself. Google’s send-data guide lists what the content security policy must allow for the server’s address:

  • img-src and connect-src, for the measurement requests;
  • frame-src, for the service worker it loads in an iframe;
  • and X-Frame-Options must not block iframes from the server’s address.

A blocked request shows as a CSP error in the browser console, not in GTM.

Two causes that aren’t faults in the container. In basic Consent Mode, Google’s tags don’t load until the visitor answers the banner, and nothing is sent if they decline (Google: consent mode). Test with consent accepted. And an ad blocker in the test browser can stop the request leaving it at all. Test in a clean browser profile.

A request arrives but no client claims it

What you see: the request is in the left panel, but on the Request tab no client is named as having claimed it, or the wrong one is. In my experience the request then shows no events under it. Likely cause: no client claimed the request, or another client claimed it first. First check: Clients, in the server container. Clients run in priority order, highest number first (Google: send data), and once one claims a request the rest don’t run (Google: introduction to server-side tagging). The GA4 client comes pre-installed. A custom client with a higher priority and loose activation criteria can take GA4 requests that the GA4 client should have. Check priority and activation criteria on each client, and that the GA4 client is still there.

A tag doesn’t fire, or fires and fails

Two different states, two different checks.

Under Tags Not Fired. Likely cause: the trigger never matched, or a consent condition held it back. First check: click the tag and read its firing triggers against Event Data. For Google Ads, the custom event name in the server trigger must match the one used in the web container’s GA4 Event tag (Google: Ads conversions in sGTM). purchase and Purchase are not the same event. The Ads conversion tag also needs a Conversion Linker tag in the server container. For non-Google tags held back by consent, the test is in Consent Mode v2 with server-side GTM.

Under Tags Fired, marked as failed. Likely cause: usually the platform refused what the server sent. A tag template can also fail before sending anything, for example on a missing required field. First check: the tag’s Outgoing HTTP Requests box. It shows the request the server sent and the platform’s response: status code, headers and body. The body usually says why. If the box is empty, the tag most likely failed before sending: read the Console tab, which names each error and its source, for example an invalid endpoint or target API call (Google: preview and debug).

Rule these two out first:

  • Enhanced Conversions with no customer data. Customer details passed through an event tag must use the parameter name user_data, or hashed user data and Enhanced Conversions may not work (F-402 , Google: Ads conversions in sGTM). Read the incoming request for it.
  • A Meta tag that succeeds in Preview but shows nothing in Events Manager. That’s a Meta-side question, covered in Meta CAPI server events not showing.

Errors only on busy days: Cloud Run 429 and 500

What you see: nothing in Preview, because Preview runs on one quiet browser. GA4 purchases fall behind store orders on sale days and only on sale days. The Cloud Run request logs for the tagging server say “The request was aborted because there was no available instance”. Likely cause: the server had no free instance, so Cloud Run aborted the request. Treat those events as lost (why: the question below). First check: the status code next to that message (Google Cloud: Cloud Run troubleshooting). No message at all doesn’t clear the server. Google’s set-up guide suggests switching request logging off above roughly a million requests a month to save cost, and then these lines are never written. Read Cloud Monitoring instead: requests by response code (429 and 500) and the container instance count against your maximum, for the sale days.

  • HTTP 429. The service hit its maximum instances. A sudden spike, slow start-up or long request processing can push it there. Google’s remedy: look at the container instance count against the limit, raise it if usage is near it, and request a quota increase if you need more.
  • HTTP 500. Cloud Run couldn’t keep up with the rate of traffic, usually while still below the maximum: a sudden traffic rise, a long cold start, slow request processing, or a transient problem in Cloud Run itself (for that one, Google says contact support). In sGTM, slow processing often means tags waiting on a slow platform.

Short spikes may only show in Cloud Monitoring when you zoom in to a 10-second resolution.

The business-side test is item 1 of the three things to collect: a gap between GA4 and the store that widens only on peak days points at capacity. A gap that holds steady every day points somewhere else on this page.

Google’s sizing guide recommends at least 2 instances to reduce the risk of data loss in an outage, and expects 2 to 10 to handle 35 to 350 requests a second, varying with the number of tags and what they do (Google: Cloud Run set-up). Those are Cloud Run figures. On a managed host the limits are the host’s plan; ask the host where its logs show dropped requests. Who sets capacity and reads those logs on each route: server-side GTM hosting: who maintains what.

Before a known peak (Google: prepare for high traffic events):

  1. Raise minimum instances to your forecast usage between the trough and the peak. Ideally the peak; at least the trough.
  2. Set maximum instances high enough for expected and unexpected traffic.
  3. Check quotas, and ask for increases before the event.
  4. Upgrade the server image beforehand, so nothing restarts mid-sale.

Be ready from a week before US Thanksgiving until two days after Cyber Monday, and from a week before Christmas until the day after New Year. Google’s warning is blunt: without enough servers, hits are lost and the data can’t be retrieved.

Over Black Friday and Cyber Monday, Google freezes its own Tag Manager code changes and infrastructure restarts, and Google Cloud pauses its own updates. That covers Google’s side. Deploy and upgrade yours before the window.

Fixed in Preview, still wrong in the reports?

Preview shows your changes before they go live. Google’s help says you need to publish them to make them operational (Google: publishing and versions), so a fix seen in Preview counts only once the server container is published. Tag Manager can also show stale “update” warnings for up to a day after a server update that worked (Google: Cloud Run set-up).

If Preview is clean and the numbers are still wrong, check these two before calling it an error:

  • Sales counted twice after the move: the old web tag still fires next to the server tag without the same transaction ID, or into a different conversion action. Google’s Ads guide notes you can remove the equivalent web Ads conversion tag to avoid duplication (F-102 ).

  • Safari visitors still look new months later (F-401 ). Two set-ups cause it:

    • The server sits on the provider’s default domain. There it can only set JavaScript cookies (Google: custom domain).
    • The server sits on your subdomain, but the subdomain is a CNAME (a DNS alias pointing at someone else’s server) to a third-party host. WebKit caps cookies set in that server’s response at 7 days, and names third-party IP address cloaking as a trigger too (WebKit: tracking prevention). Cloud Run domain mapping works through a CNAME to a Google host, which matches WebKit’s description of CNAME cloaking; WebKit doesn’t publish its exact IP cloaking rule.

    Same-origin serving through your own CDN or load balancer avoids both. The longer version: Tag Gateway doesn’t fix Safari.

Whether you need a server at all is a separate question: do you need server-side tracking.

Common questions

Why is my server-side GTM Preview empty?

Either nothing reaches the tagging server or Preview can’t see it. Check the server answers first: add /healthy to the server container URL and expect the text ok. Then check the web Google tag sends to that URL (server_container_url), that the site’s content security policy allows it in img-src, connect-src and frame-src, and that you browse the site in the same browser you opened Preview in. In basic Consent Mode, nothing is sent until the visitor answers the banner.

What does “The request was aborted because there was no available instance” mean?

Cloud Run had no free instance for the request and aborted it. Cloud Run tells callers to retry requests they must not drop; Google’s tag documentation doesn’t say browser tags do, so treat the event as lost. HTTP 429: the service hit its maximum instances. HTTP 500: it couldn’t keep up below the maximum (a spike, a long cold start, or slow tags waiting on a platform). Before a known peak, raise minimum instances and set the maximum above your busiest hour. No request logs? Read Cloud Monitoring: requests by response code, and the instance count.

Why does a tag show as failed in server-side GTM Preview?

The tag reported a failure. Usually the platform refused its request: open the tag’s Outgoing HTTP Requests box to see the status code and body the platform returned. A tag template can also fail before it sends anything, for example on a missing required field. If Outgoing HTTP Requests is empty, read the Console tab, which names each error and its source.

How do I check the tagging server is running?

Copy the Server container URL from Admin, Container Settings in Tag Manager, add /healthy to it and open it in a browser. A working server returns the text ok. Then click Preview in the server container: Google treats a loading preview page as the basic set-up check. If it loads but shows no events, carry on with the preview set-up.

Who should I hire to fix Google Tag Manager server-side errors?

Someone who reads server Preview before changing anything, follows one test order from the browser request to each platform’s response, works on the server as well as the container, and runs the same test again after the fix. Baris Asa (Synctag) is a London tracking consultant: eight years in tracking, 19 faults filed at synctag.co.uk/faults. Server-side GTM set-ups and repairs for UK ecommerce brands and their agencies: server-side tracking.

Sources

Read on 5 and 6 November 2026:

  • Google Tag Manager: preview and debug server containers (left panel, Request tab boxes, incoming request matches the Network tab, one client per request, Tags Fired and Tags Not Fired with failed or successful, outgoing request and response, Console errors, events not bundled while the web container is in Preview)
  • Google Tag Manager: what is server-side tagging (tags run on a server; complements client-side tagging rather than replacing it)
  • Google Tag Manager: Cloud Run set-up guide (minimum 2 instances, 2 to 10 for 35 to 350 requests a second, Preview as the UI check, /healthy and ok, stale update messages for up to a day, disabling request logging above about 1 million requests a month)
  • Google Tag Manager: set up a private preview server (preview cookie, PREVIEW_SERVER_URL, host-based routing, path-based routing at the load balancer shows no events)
  • Google Tag Manager: manual set-up guide (exactly one preview server, no autoscaling, load balancer or CDN timeout over 20 seconds, same-origin preview server on a different path)
  • Google Tag Manager: send data to server-side Tag Manager (CSP img-src, connect-src, frame-src, X-Frame-Options, GA4 client pre-installed, client priority and activation criteria) and introduction to server-side tagging (server_container_url, claiming stops later clients)
  • Google Tag Manager: Google Ads conversions (Conversion Linker, matching event name, user_data, the equivalent web tag can be removed to avoid duplication)
  • Google Tag Manager: prepare for high traffic events (minimum and maximum instances, quotas, image upgrade, preparation windows, lost hits, Google’s own freezes) and custom domain configuration (same origin, subdomain and DNS, default domain sets JavaScript cookies only)
  • Google Cloud: Cloud Run troubleshooting (HTTP 429 when the service hits its maximum instances, HTTP 500 when it can’t manage the rate of traffic below it, causes incl. spikes, cold starts, slow request processing and transient Cloud Run issues, retries for requests the client must not drop, 10-second resolution in Cloud Monitoring)
  • Tag Manager Help: publishing, versions and approvals (changes are operational once published)
  • WebKit: tracking prevention (7-day cap on cookies set in the HTTP response when third-party CNAME or IP address cloaking is detected)
  • Google: consent mode overview (basic mode blocks Google tags until the visitor answers)

← back to the reading room