[HAWQ-1486] PANIC accessing PXF HDFS table

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.3.0.0-incubating
• Component/s: External Tables, PXF
• Assignee: Vineet Goel
• Reporter: John Gaskin
• Created: 14/Jun/17 22:30
• Updated: 17/Jun/17 00:08
• Resolved: 15/Jun/17 19:19

Description

This code doesn't catch the case when churl_init_download() returns NULL. This seems to trigger a segfault at libcurl level.

Looks like we failed to connect to PXF (?).

Piece of code in HAWQ handling cUrl calls (pxfutils.c):

 100 static void process_request(ClientContext* client_context, char *uri)
 101 {
 102     size_t n = 0;
 103     char buffer[RAW_BUF_SIZE];
 104
 105     print_http_headers(client_context->http_headers);
 106     client_context->handle = churl_init_download(uri, client_context->http_headers);
 107     memset(buffer, 0, RAW_BUF_SIZE);
 108     resetStringInfo(&(client_context->the_rest_buf));
 109
 110     /*
 111      * This try-catch ensures that in case of an exception during the "communication with PXF and the accumulation of
 112      * PXF data in client_context->the_rest_buf", we still get to terminate the libcurl connection nicely and avoid
 113      * leaving the PXF server connection hung.
 114      */
 115     PG_TRY();
 116     {
 117         /* read some bytes to make sure the connection is established */
 118         churl_read_check_connectivity(client_context->handle);

Issue Links

• links to: GitHub Pull Request #1255