3.0-dev8 runtime API "clear map" not clearing the whole map about haproxy

[capflam]

Pushed ! It should be fixed now. Thanks !

[felipewd]

To make things easy to reproduce, we can load the map referencing the config-file, reload haproxy and try and clear the map right after a usual reload.

No need to use add map to reproduce the issue

[Darlelet]

Thanks for the report @felipewd!

static int cli_io_handler_clear_map(struct appctx *appctx)
{
        struct show_map_ctx *ctx = appctx->svcctx;
        int finished;

        HA_RWLOCK_WRLOCK(PATREF_LOCK, &ctx->ref->lock);
        finished = pat_ref_purge_range(ctx->ref, ctx->curr_gen, ctx->prev_gen, 100);
        HA_RWLOCK_WRUNLOCK(PATREF_LOCK, &ctx->ref->lock);

        if (!finished) {
                /* let's come back later */
                applet_have_more_data(appctx);

map pruning is segmented through batches of 100 entries each to prevent the process from monopolizing a CPU core which would cause contention otherwise. If entries still remain after a pruning batch, the applet is re-scheduled to continue where it left off.

As you pointed out this used to work properly in the past, but since f37ddbe ("MAJOR: cli: Update the CLI applet to handle its own buffers") it looks like the applet is not woken back properly and because of that it doesn't get the chance to finish the pruning process, and only a few hundreds items maximum are removed. When this happens, we can also notice that the cli command hangs instead of returning promptly.

[capflam]

Indeed. Thanks @Darlelet ! The issue is in sc_applet_recv(). SE_FL_HAVE_NO_DATA flag is updated in the same way than for muxes while it is in fact the applet responsibility to do so. I will push a commit soon. But in the mean time the following should fix the issue:

diff --git a/src/stconn.c b/src/stconn.c
index 930b92b1bd..621d51e429 100644
--- a/src/stconn.c
+++ b/src/stconn.c
@@ -2094,9 +2094,7 @@ int sc_applet_recv(struct stconn *sc)
        }
 
  done_recv:
-       if (!cur_read)
-               se_have_no_more_data(sc->sedesc);
-       else {
+       if (cur_read) {
                channel_check_xfer(ic, cur_read);
                sc_ep_report_read_activity(sc);
        }
@@ -2125,12 +2123,7 @@ int sc_applet_recv(struct stconn *sc)
                sc->flags |= SC_FL_ERROR;
                ret = 1;
        }
-       else if (!cur_read &&
-                !(sc->flags & (SC_FL_WONT_READ|SC_FL_NEED_BUFF|SC_FL_NEED_ROOM)) &&
-                !(sc->flags & (SC_FL_EOS|SC_FL_ABRT_DONE))) {
-               se_have_no_more_data(sc->sedesc);
-       }
-       else {
+       else if (cur_read || (sc->flags & (SC_FL_WONT_READ|SC_FL_NEED_BUFF|SC_FL_NEED_ROOM))) {
                se_have_more_data(sc->sedesc);
                ret = 1;
        }

[felipewd]

hey @capflam I can confirm this patches fixes the issue, thanks!