Hi, Are you sure it is only the first write? I’m not aware of any lazy initialization there, so I wouldn’t expect the first write to be slow.
Which version do you use? And what is the performance difference of the first transaction compared to the second? I have two ideas why it may work slowly on the first write 1) WAL roll-over – every time you start to write to a new WAL segment it will initialize it first; perhaps it is done synchronously for the first segment. If it is the reason then the next roll-over will also be slow and cause timeout. Try to check if there are periodic slow transactions. 2) Usual JVM warmup – JVM may need to load some classes, JIT simple methods, etc. I guess -Xcompile would help with JIT here - it will force JVM to JIT methods before the first execution. Although it seems unlikely that the performance difference would be so big that it causes timeouts. Thanks, Stan From: Andrey Davydov Sent: 12 декабря 2018 г. 19:36 To: user@ignite.apache.org Subject: First write to new cache on new cluster. Hello all, The first write transaction on empty cluster with persistence enabled takes some seconds. After that, cluster works fast. It seems it is because of WAL file creation and so on. But I can’t setup real transaction timeout in configuration, because it cause exception on the first operation. I can do dummy transaction to force initialization. But it seems not best practice. If there is convenient way to force persistence initialization? Andrey.
