Brakes for your Cloudflare bill.

Queue retry storms

Consumers retrying the same messages over and over, or set to retry many times with no dead letter queue.

The fix Keep max_retries low and add a dead_letter_queue, so failures stop and stay visible.

              
                  RUNAWAY
                   
                  Messages are being retried over and over
                
                  
                  
                  email-queue
                
                  
                  
                  300,000 retries for 100,000 messages (3.00 per message)
                
            

Crons that grow with your data

Frequent jobs with heavy CPU, and runs creeping toward the 15-minute limit, where Cloudflare reports per-cron usage.

The fix Process a bounded batch per run, with a cursor saved between runs.

              
                  RUNAWAY
                   
                  Cron work grows with your data
                
                  
                  
                  sync-worker (* * * * *)
                
                  
                  
                  CPU per run grew 3.0x across the window (10,000 to 30,000 ms)
                
            

D1 queries with no index

Reads and writes that scan a whole table for a handful of rows. D1 bills every row scanned.

The fix Add an index on the columns in its WHERE clause, then check the plan with EXPLAIN QUERY PLAN.

              
                  WARNING
                   
                  A query reads far more rows than it returns
                
                  
                  
                  prod-db
                
                  
                  
                  50,000 rows read for every row it returns
                
            

Data with no outside copy

A reminder for every D1 database. Copies you keep elsewhere can't be seen from Cloudflare's API, so all of them are listed.

The fix Export it on a schedule to storage you control, and test a restore.

              
                  NOTE
                   
                  No copy of this data outside Cloudflare
                
                  
                  
                  prod-db
                
                  
                  
                  2.50 GB stored, and no outside copy was reported
                
            

Durable Object alarm loops

Cloudflare's analytics don't report alarms separately, so a usage scan can't tell an alarm loop from ordinary traffic. The report says so, instead of quietly showing zero.

              
                  What this scan could not measure
                  
                  
                
                  
                  
                  - Cloudflare's analytics don't report Durable Object alarms separately, so alarm loops can't be detected from usage yet.