Please sign in to comment.
Chart repo index fetcher is moved to a background job
Due to a massive overhead that happened as a consequence of a heavy chart repo index request pattern (introduced in 0.5.0), application and installation controller latencies spiked up. This change is aiming to get rid of ad-hoc repo index fetch approach and move this heavy operation to a background job. From now on, every repo instance start polling chart repo index every 10 seconds. Once the data is successfully fetched, it is preserved as a repo attribute unmarshalled. The first fetch is blocking: assuming Shipper starts cold, there is no previous cache we can rely upon (it starts in a new container). On top of it, index data is never cached on the disk as there is no use for it any longer: in-memory only. If repo fails to fetch repo index, it behaves quite naively: simply spins next iteration with the same delay. Signed-off-by: Oleg Sidorov <email@example.com>
- Loading branch information...
Showing with 80 additions and 70 deletions.