[HDDS-6364] EC: Discard pre-allocated blocks to eliminate worthless retries.

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: EC-Branch
• Labels: pull-request-available
• Assignee: Mark Gui
• Reporter: Mark Gui
• Created: 23/Feb/22 03:08
• Updated: 01/Mar/22 19:21
• Resolved: 01/Mar/22 19:21

Description

When we meet failures during a stripe write, we'll rewrite the stripe into a new block group, this new block group should be from a different pipeline with some different DNs to ensure that we have a good chance to succeed at the next retry.

But there are pre-allocated block groups when we create a new EC key, for a large EC key there will be at most 64 pre-allocated blocks(ozone.key.preallocation.max.blocks = 64, by default), and we've only got 10 retries(ozone.client.max.ec.stripe.write.retries = 10) for EC stripe write, so the pre-allocated blocks came from the same pipeline as the failed block group should be discarded to prevent worthless retries.

Issue Links

• links to: GitHub Pull Request #3127