Laurent, I don't have answers to all your questions, but we use Glorp & Seaside (no Magritte) on VA Smalltalk for our cloud service (http://kontolino.de). Am 18.09.17 um 12:22 schrieb laurent:
Hi all,
it's been a long time :) At work we are currently comparing several technologies to start a business project that will hopefully be used by thousands of people for several years ;)
We need a web stack and we put Pharo & co in the comparative process. My team have strong experience in  load-balanced php / mysql deployment that handles millions of records  but we are ready to build on something else though MySQL is still a requirement. Sounds interesting.
Thanks for Garage + Glorp, I can prototype things. I wonder how to handle database schema migrations. With our PHP projects we have versionned PHP scripts that are automatically run given a database revision number ( see http://git.afi-sa.fr/afi/opacce/tree/master/cosmogramme/sql/patch ). What are the practices with Glorp ?
We use a similar approach: our Descriptor class has a method called schemaVersion, which just returns the current version number. We also have a "magic" table for settings in our DB, just holding a "name" and a "value" column, both being Varchars. One of the rows has a name of "SchemaVersion" and the value is (you guessed it) the printString of the version number. Our images accept a few command line paramters, one of them being --upgradeSchema. It compoares the number in the DB with the one provided by the descriptor. If they match, the image quits. If not, it calls all methods named #upgradeSchemaToVersionXXX until it reaches the number in the Descriptor. Thus we can do both SQL based migrations, like DDL stuff, as well as Smalltalk object based migrations (which can be sooo much faster to implement for complex cases). And we have all versioned and maintained in our beloved Envy repository ;-) . The rest is just bash script magic (like only starting one image with the --upgradeSchema parameter and only start the real web workers once that one has ended succesfully). About load balancing: You need Sticky sessions, so no failover between images. If one image crashes, the user must re-logon on a new image. Works nicely with Apache, I guess nginx works as least as well. For Apache, look for mod_prox_balancer to learn more. Hope I could give you a few starting points for your research. And I also hope you'll find Pharo+Garage+Glorp+QCMagritte a good fit for your project. We all want success stories ;-) Joachim -- ----------------------------------------------------------------------- Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de Fliederweg 1Â Â Â Â Â Â Â Â Â Â Â Â Â http://www.objektfabrik.de D-71640 Ludwigsburg http://joachimtuchel.wordpress.com Telefon: +49 7141 56 10 86 0Â Â Â Â Â Fax: +49 7141 56 10 86 1