Why do support dates drive migrations?
Because they turn a deferred decision into a dated one. A warehouse that works can run for years without a business case to move it. An end-of-support date changes that. After it, the vendor stops shipping security fixes, auditors start asking questions, and insurers and regulators may follow.
The date also limits the options. A team that starts counting its logic three months before the date has fewer choices than one that starts eighteen months before. This note collects the published dates so the planning can start early.
Method. Dates come from Microsoft’s and Cloudera’s published lifecycle pages, checked on 11 November 2026. Vendors sometimes change their policies, so check the linked pages before relying on a date.
When does support end for SQL Server?
Microsoft publishes a fixed lifecycle for each SQL Server version: mainstream support, then extended support, then in some cases paid Extended Security Updates (ESUs). Mainstream support includes feature and non-security fixes. Extended support provides security updates only.
| Version | Mainstream support ended | Extended support ends | What it means |
|---|---|---|---|
| SQL Server 2016 | Ended | July 2026 (ended); Microsoft lists ESUs to July 2029 | Past extended support. Security fixes now need paid ESUs. |
| SQL Server 2017 | Ended | October 2027 | About eleven months of extended support left from this note’s date. |
| SQL Server 2019 | February 2025 | January 2030 | In extended support: security fixes only. SQL Server Big Data Clusters were retired in February 2025. |
Dates from Microsoft Learn lifecycle pages. Microsoft lists SQL Server 2019’s extended end date as 9 January 2030, 06:59 Pacific Time, so support ends at the end of 8 January 2030 US time.
When does support end for Hadoop distributions?
Most on-premises Hadoop estates run a commercial distribution from Cloudera, which merged with Hortonworks in 2019. Cloudera publishes support dates by version.
| Distribution and version | End of support | Notes |
|---|---|---|
| HDP 3.1 (Hortonworks Data Platform) | December 2021 | Past end of support |
| Cloudera Enterprise 6.3 (CDH 6.3) | March 2022 | Past end of support; limited support ran to 30 September 2022 for eligible customers |
| CDP Private Cloud Base 7.1.7 | August 2025 | Past end of support |
| CDP Private Cloud Base 7.1.8 | August 2024 | Past end of support |
| CDP Private Cloud Base 7.1.9 | October 2028 | Supported; Cloudera states end of support falls on the last day of the month listed |
Dates from Cloudera’s published support lifecycle policy. Estates on CDH 6 or HDP 3 are already unsupported; the Hive and Impala logic on them still has to move.
What do Extended Security Updates buy?
Time, not progress. ESUs keep security fixes coming for a fee after extended support ends. They do not move a single stored procedure, and they do not reduce the logic that has to be converted later.
ESUs are a reasonable bridge for a team that has already counted its estate, chosen its target and started moving domains. They are an expensive way to postpone a decision for a team that has not.
How should a team plan backwards from the date?
Start from the date and work back. The timeline below assumes a mid-size estate of a few thousand objects; larger estates need more time for the later waves.
| Months before the date | What should be done | Why |
|---|---|---|
| 18 | Count every object on the platform, with dead code and complexity bands | Without a count, every option is priced on a guess |
| 15 | Choose the target platform and the migration method, with a fixed price | The method decides the timeline, and the price decides the budget |
| 12 | Move and prove the first self-contained domain | It tests the method on your own code before the estate is committed |
| 9 to 3 | Move the remaining domains in waves, each proven before cutover | Domain by domain keeps each parallel run short |
| 3 to 0 | Final cutover, then retire the old platform | Leave a margin for the domain that surprises you |
What else sets a date?
Support dates are not the only trigger. License renewals for platforms such as Teradata and Snowflake give a date and a price, and the useful question is what it would cost to leave, fixed, before you sign. Platform mandates, such as a company decision to standardize on Microsoft Fabric or Databricks, set a date through budget and licensing rather than support. Hardware refreshes do the same for on-premises Hadoop clusters.
Each works like a support date: count the logic first, then decide with two numbers instead of one.
What are the limits of this note?
Dates are as published on 11 November 2026 and can change. Vendors sometimes extend support or change their ESU terms. Always check the vendor’s lifecycle page for your exact version and edition.
The planning timeline is a starting point, not a rule. It depends on the size of the estate, how much of it is dead code, and how quickly domain owners can test.