Pulsar - Transaction Message Detail Question

[yutao chen <ch...@gmail.com>]

Hi, I'm very interested in Pulsar transaction message, and I found this
issue <https://github.com/apache/pulsar/pull/8347>
Would you please tell me why pulsar finally chose to replace sidecar's
appraoch with marker's appraoch, which seems different from the conclusion
in the previous PIP-31
<https://docs.google.com/document/d/145VYp09JKTw9jAT-7yNyFU255FptB2_B2Fye100ZXDI/edit#>

Re: Pulsar - Transaction Message Detail Question

[PengHui Li <pe...@apache.org>]

Hi Yutao,

Thanks for your feedback.

Yes, the current implementation is the marker approach.
And the transaction buffer is an interface, in fact, this
does not limit we can only have one approach.

When we implement the sidecar approach, the biggest
challenge is mapping the message ID to the entry of the
transaction commit marker, as described in the proposal,
we need an extra index to make it work and the sidecar
approach will also double the number of managed ledgers,
which will bring more burden to Zookeeper.

There are indeed some details that we did not update back
to the proposal, we will prepare this part to update it back to PIP.

Thanks,
Penghui

On Mon, Feb 21, 2022 at 11:51 PM yutao chen <ch...@gmail.com> wrote:

> Hi, I'm very interested in Pulsar transaction message, and I found this
> issue <https://github.com/apache/pulsar/pull/8347>
> Would you please tell me why pulsar finally chose to replace sidecar's
> appraoch with marker's appraoch, which seems different from the conclusion
> in the previous PIP-31
> <
> https://docs.google.com/document/d/145VYp09JKTw9jAT-7yNyFU255FptB2_B2Fye100ZXDI/edit#
> >
>