Container Registry SLA Credits: What ECR, ACR and DigitalOcean Pay When Pulls Fail
October 4, 2026
On 20 October 2025, Docker Hub went dark for eleven hours after the AWS us-east-1 failure. Pulls failed worldwide; Docker published a post-mortem and zero compensation, because Docker Hub has no uptime SLA. No credit tier, no claim window, nothing to file.
That is not a Docker quirk. Each provider treats its registry differently. ECR promises 99.9% with a 10/25/100 ladder and a $1 floor. Azure Container Registry promises 99.9% but caps credits at 25% and forgives slow-but-alive registries. DigitalOcean promises 99.95% but starts the downtime clock only when you report. Google's Artifact Registry, the only registry left after gcr.io shut down in March 2025, has no SLA page at all. Here is who owes what, what you must prove, and what the credit is worth.
The registry SLA map
| Registry | Commitment | How downtime is measured | Credit tiers | Claim window | Minimum credit |
|---|---|---|---|---|---|
| Amazon ECR | 99.9% per region | 5-minute request intervals; requests failing with ECR internal errors; empty intervals count as fully available | 10% / 25% / 100% | End of second billing cycle after the incident | $1 |
| Azure Container Registry | 99.9% (Basic, Standard, Premium) | Minutes where all transaction attempts error or exceed max processing time: 8 minutes for List operations, 1 minute for other operations | 10% / 25% only | 60 days from the incident | None stated |
| DigitalOcean Container Registry | 99.95% per billing cycle | 5-minute intervals with no external accessibility; clock starts when you report or acknowledge | 10% / 25% / 100% | 30 days after end of billing cycle | $1 |
| Google Artifact Registry | None | No SLA exists | None | n/a | n/a |
| Docker Hub | None | Support SLA only | None | n/a | n/a |
Every paid credit is a percentage of your registry bill, never your compute bill. And the two registries most teams touch first, Artifact Registry and Docker Hub, pay nothing.
Amazon ECR: a claim you file with logs
ECR's SLA, May 4 2022, promises 99.9% Monthly Uptime Percentage per region. Availability is the share of requests per 5-minute interval that do not fail with an internal ECR error. Intervals with no requests count as 100% available; your network, client bugs and credentials are excluded.
Credits are the standard AWS ladder: 10% below 99.9%, 25% below 99.0%, 100% below 95.0%, applied to your ECR charges in that region for the month. Credits issue only above one dollar, cannot transfer between accounts, and are your sole remedy.
To file: open an AWS Support case with "SLA Credit Request" in the subject, the billing cycle and region, the Monthly Uptime Percentage, the dates of every 5-minute interval under 100%, and request logs showing the errors, by the end of the second billing cycle after the incident. The evidence checklist lists what support accepts.
The credit base: ECR charges $0.10 per GB-month of storage; in-region pulls are free. A team storing 40 GB pays about $4 a month, and a 25% tier breach on that bill is one dollar, exactly at the floor, so it pays nothing.
Azure Container Registry: transaction minutes, no 100% tier
The October 2026 consolidated SLA covers Managed Registries, defined as Basic, Standard or Premium. The commitment is 99.9%, measured through transactions: a minute counts as downtime only if all attempts to send Registry Transactions error or exceed the Maximum Processing Time, 8 minutes for List operations and 1 minute for everything else.
The clock forgives a lot: a slow registry answering a fraction of requests stays "available" because the minute fails only when every attempt fails.
Credits are the Azure 10/25 shape with no 100% tier, applied to the Applicable Service Fees actually paid in the period around the incident. Claims go in through an Azure support request within 60 days of the incident.
Azure pricing keeps the math small. Basic runs about $5 a month with 10 GB included, Standard about $20 with 100 GB, Premium about $50 with 500 GB, plus $0.10 per GB-month beyond. A 15-hour outage puts the month at 97.9% uptime, the 25% tier: $1.25 on Basic, $5 on Standard.
DigitalOcean: the clock starts when you call
DigitalOcean's Container Registry SLA, December 2025, promises 99.95% per billing cycle: the share of 5-minute intervals in which every image has external accessibility for push and pull. The tightest commitment of the three, with the strangest clock:
Downtime will begin to accrue as soon as you, subject to DigitalOcean's confirmation, either report that an outage is taking place or acknowledge a pro-active report from DigitalOcean.
DigitalOcean does not backdate. A registry unreachable from 02:00 to 04:00 does not count unless you reported it, or acknowledged their report, at the time. The ticket must go in before the interval math starts.
Exclusions cover maintenance, force majeure, quota limits and "data loss (including Resources)". Credits are 10/25/100 on Container Registry charges, must clear the $1 floor, and expire after 90 days. Request them within 30 days of the end of the billing cycle. Pricing: Starter $0, Basic $5, Professional $20, overage $0.02 per GiB.
The two registries that pay nothing
Artifact Registry is not in Google's SLA repository. The list covers more than a hundred services and Artifact Registry is absent; the SLA URL returns a 404. After Container Registry shut down on March 18 2025, Artifact Registry is the only Google registry, and Google's credit system never starts for it. A week-long outage on GCP earns exactly zero.
Docker's trust page says the Hub "consistently achieves 99.9% uptime", a marketing claim, not a commitment; the only SLA on docker.com covers support response times. Rate limits are not an outage: anonymous pulls cap at 100 per 6 hours per IP, Personal at 200, paid plans unlimited. A 429 is the product working as designed.
What a claim is actually worth
A 15-hour outage in a 30-day month is 97.9% uptime: the 25% tier on all three paid registries.
| Registry | Monthly bill | Credit at 25% | Paid out |
|---|---|---|---|
| ECR, 40 GB storage | $4.00 | $1.00, exactly at the floor | $0 |
| ECR, 100 GB storage | $10.00 | $2.50 | $2.50 |
| ACR Basic | $5.00 | $1.25 | $1.25 |
| ACR Standard | $20.00 | $5.00 | $5.00 |
| DigitalOcean Basic | $5.00 | $1.25 | $1.25 |
| DigitalOcean Professional | $20.00 | $5.00 | $5.00 |
| Artifact Registry | any | none, no SLA | $0 |
| Docker Hub, any paid plan | any | none, no SLA | $0 |
Registry SLAs are the smallest credits in cloud computing: a 15-hour outage on a $5 registry pays about $1.25, and on Artifact Registry or Docker Hub it pays exactly zero because there is no SLA to breach. The resilience answer is a cached copy, not a claim form.
What to do
If a registry outage hit you in the last two billing cycles: ECR closes at the end of the second billing cycle after the incident, Azure gives 60 days, DigitalOcean 30 days from month end. ECR evidence is logs, Azure the support-request narrative, DigitalOcean your live report record: start a ticket while the incident is open.
Treat the credit as the rounding error it is. Point your clusters and CI at a mirror of the images you deploy: a pull-through cache or a nightly copy into a second registry costs a few dollars a month and turns the next eleven-hour registry outage into a non-event. On 20 October 2025 the teams with a cached copy kept deploying; the rest got an apology.